[WIP] Handling binary data in DL4J servlet (#135)

* Binary deser

Signed-off-by: Alexander Stoyakin <alexander.stoyakin@gmail.com>

* Binary mode for servlet

Signed-off-by: Alexander Stoyakin <alexander.stoyakin@gmail.com>

* Added test

Signed-off-by: Alexander Stoyakin <alexander.stoyakin@gmail.com>

* -sRandom image generation copied from datavec

* -sRandom image generation copied from datavec

* Remove serialization constraints

* Fix:

Signed-off-by: Alexander Stoyakin <alexander.stoyakin@gmail.com>

* Removed unused code

Signed-off-by: Alexander Stoyakin <alexander.stoyakin@gmail.com>

* Resources usage

Signed-off-by: Alexander Stoyakin <alexander.stoyakin@gmail.com>

* Async inference

Signed-off-by: Alexander Stoyakin <alexander.stoyakin@gmail.com>

* Cleanup

Signed-off-by: Alexander Stoyakin <alexander.stoyakin@gmail.com>

* -sTest corrected

* Cleanup

Signed-off-by: Alexander Stoyakin <alexander.stoyakin@gmail.com>

* Mutually eclusive serializers/deserializers

Signed-off-by: Alexander Stoyakin <alexander.stoyakin@gmail.com>

* Binary output supported

Signed-off-by: Alexander Stoyakin <alexander.stoyakin@gmail.com>

* Binary out test

Signed-off-by: Alexander Stoyakin <alexander.stoyakin@gmail.com>

* - types hardcoded
- increased payload size limit

Signed-off-by: raver119 <raver119@gmail.com>

* change types constant

Signed-off-by: raver119 <raver119@gmail.com>

deeplearning4j/deeplearning4j-remote/deeplearning4j-json-server/src/main/java/org/deeplearning4j/remote/DL4jServlet.java

@@ -18,18 +18,17 @@ package org.deeplearning4j.remote;
 
 import lombok.*;
 import lombok.extern.slf4j.Slf4j;
-import org.deeplearning4j.nn.api.Layer;
 import org.deeplearning4j.nn.api.Model;
-import org.deeplearning4j.nn.api.NeuralNetwork;
 import org.deeplearning4j.nn.graph.ComputationGraph;
 import org.deeplearning4j.nn.multilayer.MultiLayerNetwork;
 import org.deeplearning4j.parallelism.ParallelInference;
 import org.nd4j.adapters.InferenceAdapter;
 import org.nd4j.base.Preconditions;
-import org.nd4j.linalg.api.ndarray.INDArray;
+import org.nd4j.linalg.dataset.MultiDataSet;
+import org.nd4j.remote.clients.serde.BinaryDeserializer;
+import org.nd4j.remote.clients.serde.BinarySerializer;
 import org.nd4j.remote.clients.serde.JsonDeserializer;
 import org.nd4j.remote.clients.serde.JsonSerializer;
-import org.nd4j.adapters.InferenceAdapter;
 import org.nd4j.remote.serving.SameDiffServlet;
 
 import javax.servlet.http.HttpServletRequest;
@@ -38,6 +37,8 @@ import java.io.BufferedReader;
 import java.io.IOException;
 import java.io.InputStreamReader;
 
+
+
 /**
  *
  * @author astoyakin
@@ -51,7 +52,7 @@ public class DL4jServlet<I,O> extends SameDiffServlet<I,O> {
     protected boolean parallelEnabled = true;
 
     public DL4jServlet(@NonNull ParallelInference parallelInference, @NonNull InferenceAdapter<I, O> inferenceAdapter,
-                       @NonNull JsonSerializer<O> serializer, @NonNull JsonDeserializer<I> deserializer) {
+                       JsonSerializer<O> serializer, JsonDeserializer<I> deserializer) {
         super(inferenceAdapter, serializer, deserializer);
         this.parallelInference = parallelInference;
         this.model = null;
@@ -59,19 +60,67 @@ public class DL4jServlet<I,O> extends SameDiffServlet<I,O> {
     }
 
     public DL4jServlet(@NonNull Model model, @NonNull InferenceAdapter<I, O> inferenceAdapter,
-                       @NonNull JsonSerializer<O> serializer, @NonNull JsonDeserializer<I> deserializer) {
+                       JsonSerializer<O> serializer, JsonDeserializer<I> deserializer) {
         super(inferenceAdapter, serializer, deserializer);
         this.model = model;
         this.parallelInference = null;
         this.parallelEnabled = false;
     }
 
+    public DL4jServlet(@NonNull ParallelInference parallelInference, @NonNull InferenceAdapter<I, O> inferenceAdapter,
+                       BinarySerializer<O> serializer, BinaryDeserializer<I> deserializer) {
+        super(inferenceAdapter, serializer, deserializer);
+        this.parallelInference = parallelInference;
+        this.model = null;
+        this.parallelEnabled = true;
+    }
+
+    public DL4jServlet(@NonNull Model model, @NonNull InferenceAdapter<I, O> inferenceAdapter,
+                       JsonSerializer<O> jsonSerializer, JsonDeserializer<I> jsonDeserializer,
+                       BinarySerializer<O> binarySerializer, BinaryDeserializer<I> binaryDeserializer) {
+        super(inferenceAdapter, jsonSerializer, jsonDeserializer, binarySerializer, binaryDeserializer);
+        this.model = model;
+        this.parallelInference = null;
+        this.parallelEnabled = false;
+    }
+
+    public DL4jServlet(@NonNull ParallelInference parallelInference, @NonNull InferenceAdapter<I, O> inferenceAdapter,
+                       JsonSerializer<O> jsonSerializer, JsonDeserializer<I> jsonDeserializer,
+                       BinarySerializer<O> binarySerializer, BinaryDeserializer<I> binaryDeserializer) {
+        super(inferenceAdapter, jsonSerializer, jsonDeserializer, binarySerializer, binaryDeserializer);
+        this.parallelInference = parallelInference;
+        this.model = null;
+        this.parallelEnabled = true;
+    }
+
+    private O process(MultiDataSet mds) {
+        O result = null;
+        if (parallelEnabled) {
+            // process result
+            result = inferenceAdapter.apply(parallelInference.output(mds.getFeatures(), mds.getFeaturesMaskArrays()));
+        } else {
+            synchronized (this) {
+                if (model instanceof ComputationGraph)
+                    result = inferenceAdapter.apply(((ComputationGraph) model).output(false, mds.getFeatures(), mds.getFeaturesMaskArrays()));
+                else if (model instanceof MultiLayerNetwork) {
+                    Preconditions.checkArgument(mds.getFeatures().length > 0 || (mds.getFeaturesMaskArrays() != null && mds.getFeaturesMaskArrays().length > 0),
+                            "Input data for MultilayerNetwork is invalid!");
+                    result = inferenceAdapter.apply(((MultiLayerNetwork) model).output(mds.getFeatures()[0], false,
+                            mds.getFeaturesMaskArrays() != null ? mds.getFeaturesMaskArrays()[0] : null, null));
+                }
+            }
+        }
+        return result;
+    }
+
     @Override
     protected void doPost(HttpServletRequest request, HttpServletResponse response) throws IOException {
         String processorReturned = "";
+        MultiDataSet mds = null;
         String path = request.getPathInfo();
         if (path.equals(SERVING_ENDPOINT)) {
             val contentType = request.getContentType();
+            if (contentType.equals(typeJson)) {
                 if (validateRequest(request, response)) {
                     val stream = request.getInputStream();
                     val bufferedReader = new BufferedReader(new InputStreamReader(stream));
@@ -83,31 +132,35 @@ public class DL4jServlet<I,O> extends SameDiffServlet<I,O> {
                     }
                     val requestString = buffer.toString();
 
-                val mds = inferenceAdapter.apply(deserializer.deserialize(requestString));
+                    mds = inferenceAdapter.apply(deserializer.deserialize(requestString));
-
+                }
-                O result = null;
+            }
-                if (parallelEnabled) {
+            else if (contentType.equals(typeBinary)) {
-                    // process result
+                val stream = request.getInputStream();
-                    result = inferenceAdapter.apply(parallelInference.output(mds.getFeatures(), mds.getFeaturesMaskArrays()));
+                int available = request.getContentLength();
+                if (available <= 0) {
+                    response.sendError(411, "Content length is unavailable");
                 }
                 else {
-                    synchronized(this) {
+                    byte[] data = new byte[available];
-                        if (model instanceof ComputationGraph)
+                    stream.read(data, 0, available);
-                            result = inferenceAdapter.apply(((ComputationGraph)model).output(false, mds.getFeatures(), mds.getFeaturesMaskArrays()));
+
-                        else if (model instanceof MultiLayerNetwork) {
+                    mds = inferenceAdapter.apply(binaryDeserializer.deserialize(data));
-                            Preconditions.checkArgument(mds.getFeatures().length > 1 || (mds.getFeaturesMaskArrays() != null && mds.getFeaturesMaskArrays().length > 1),
-                                    "Input data for MultilayerNetwork is invalid!");
-                            result = inferenceAdapter.apply(((MultiLayerNetwork) model).output(mds.getFeatures()[0], false,
-                                    mds.getFeaturesMaskArrays() != null ? mds.getFeaturesMaskArrays()[0] : null, null));
                 }
             }
+            if (mds == null)
+                log.error("InferenceAdapter failed");
+            else {
+                val result = process(mds);
+                if (binarySerializer != null) {
+                    byte[] serialized = binarySerializer.serialize(result);
+                    response.setContentType(typeBinary);
+                    response.setContentLength(serialized.length);
+                    val out = response.getOutputStream();
+                    out.write(serialized);
                 }
+                else {
                     processorReturned = serializer.serialize(result);
-            }
-        } else {
-            // we return error otherwise
-            sendError(request.getRequestURI(), response);
-        }
                     try {
                         val out = response.getWriter();
                         out.write(processorReturned);
@@ -115,6 +168,12 @@ public class DL4jServlet<I,O> extends SameDiffServlet<I,O> {
                         log.error(e.getMessage());
                     }
                 }
+            }
+        } else {
+            // we return error otherwise
+            sendError(request.getRequestURI(), response);
+        }
+    }
 
     /**
      * Creates servlet to serve models
@@ -133,6 +192,8 @@ public class DL4jServlet<I,O> extends SameDiffServlet<I,O> {
         private InferenceAdapter<I, O> inferenceAdapter;
         private JsonSerializer<O> serializer;
         private JsonDeserializer<I> deserializer;
+        private BinarySerializer<O> binarySerializer;
+        private BinaryDeserializer<I> binaryDeserializer;
         private int port;
         private boolean parallelEnabled = true;
 
@@ -155,7 +216,7 @@ public class DL4jServlet<I,O> extends SameDiffServlet<I,O> {
          * @param serializer
          * @return
          */
-        public Builder<I,O> serializer(@NonNull JsonSerializer<O> serializer) {
+        public Builder<I,O> serializer(JsonSerializer<O> serializer) {
             this.serializer = serializer;
             return this;
         }
@@ -166,11 +227,33 @@ public class DL4jServlet<I,O> extends SameDiffServlet<I,O> {
          * @param deserializer
          * @return
          */
-        public Builder<I,O> deserializer(@NonNull JsonDeserializer<I> deserializer) {
+        public Builder<I,O> deserializer(JsonDeserializer<I> deserializer) {
             this.deserializer = deserializer;
             return this;
         }
 
+        /**
+         * This method is required to specify serializer
+         *
+         * @param serializer
+         * @return
+         */
+        public Builder<I,O> binarySerializer(BinarySerializer<O> serializer) {
+            this.binarySerializer = serializer;
+            return this;
+        }
+
+        /**
+         * This method allows to specify deserializer
+         *
+         * @param deserializer
+         * @return
+         */
+        public Builder<I,O> binaryDeserializer(BinaryDeserializer<I> deserializer) {
+            this.binaryDeserializer = deserializer;
+            return this;
+        }
+
         /**
          * This method allows to specify port
          *
@@ -194,8 +277,8 @@ public class DL4jServlet<I,O> extends SameDiffServlet<I,O> {
         }
 
         public DL4jServlet<I,O> build() {
-            return parallelEnabled ? new DL4jServlet<I, O>(pi, inferenceAdapter, serializer, deserializer) :
+            return parallelEnabled ? new DL4jServlet<I, O>(pi, inferenceAdapter, serializer, deserializer, binarySerializer, binaryDeserializer) :
-                    new DL4jServlet<I, O>(model, inferenceAdapter, serializer, deserializer);
+                    new DL4jServlet<I, O>(model, inferenceAdapter,  serializer, deserializer, binarySerializer, binaryDeserializer);
         }
     }
 }

deeplearning4j/deeplearning4j-remote/deeplearning4j-json-server/src/main/java/org/deeplearning4j/remote/JsonModelServer.java

@@ -34,6 +34,8 @@ import org.nd4j.linalg.api.ndarray.INDArray;
 import org.nd4j.linalg.dataset.MultiDataSet;
 import org.nd4j.linalg.factory.Nd4j;
 import org.nd4j.remote.SameDiffJsonModelServer;
+import org.nd4j.remote.clients.serde.BinaryDeserializer;
+import org.nd4j.remote.clients.serde.BinarySerializer;
 import org.nd4j.remote.clients.serde.JsonDeserializer;
 import org.nd4j.remote.clients.serde.JsonSerializer;
 
@@ -70,28 +72,40 @@ public class JsonModelServer<I, O> extends SameDiffJsonModelServer<I, O> {
 
     protected boolean enabledParallel = true;
 
-    protected JsonModelServer(@NonNull SameDiff sdModel, InferenceAdapter<I, O> inferenceAdapter, JsonSerializer<O> serializer, JsonDeserializer<I> deserializer, int port, String[] orderedInputNodes, String[] orderedOutputNodes) {
+    protected JsonModelServer(@NonNull SameDiff sdModel, InferenceAdapter<I, O> inferenceAdapter,
-        super(sdModel, inferenceAdapter, serializer, deserializer, port, orderedInputNodes, orderedOutputNodes);
+                              JsonSerializer<O> serializer, JsonDeserializer<I> deserializer,
+                              BinarySerializer<O> binarySerializer, BinaryDeserializer<I> binaryDeserializer,
+                              int port, String[] orderedInputNodes, String[] orderedOutputNodes) {
+        super(sdModel, inferenceAdapter, serializer, deserializer, binarySerializer, binaryDeserializer, port, orderedInputNodes, orderedOutputNodes);
     }
 
-    protected JsonModelServer(@NonNull ComputationGraph cgModel, InferenceAdapter<I, O> inferenceAdapter, JsonSerializer<O> serializer, JsonDeserializer<I> deserializer, int port, @NonNull InferenceMode inferenceMode, int numWorkers) {
+    protected JsonModelServer(@NonNull ComputationGraph cgModel, InferenceAdapter<I, O> inferenceAdapter,
-        super(inferenceAdapter, serializer, deserializer, port);
+                              JsonSerializer<O> serializer, JsonDeserializer<I> deserializer,
+                              BinarySerializer<O> binarySerializer, BinaryDeserializer<I> binaryDeserializer,
+                              int port, @NonNull InferenceMode inferenceMode, int numWorkers) {
+        super(inferenceAdapter, serializer, deserializer, binarySerializer, binaryDeserializer, port);
 
         this.cgModel = cgModel;
         this.inferenceMode = inferenceMode;
         this.numWorkers = numWorkers;
     }
 
-    protected JsonModelServer(@NonNull MultiLayerNetwork mlnModel, InferenceAdapter<I, O> inferenceAdapter, JsonSerializer<O> serializer, JsonDeserializer<I> deserializer, int port, @NonNull InferenceMode inferenceMode, int numWorkers) {
+    protected JsonModelServer(@NonNull MultiLayerNetwork mlnModel, InferenceAdapter<I, O> inferenceAdapter,
-        super(inferenceAdapter, serializer, deserializer, port);
+                              JsonSerializer<O> serializer, JsonDeserializer<I> deserializer,
+                              BinarySerializer<O> binarySerializer, BinaryDeserializer<I> binaryDeserializer,
+                              int port, @NonNull InferenceMode inferenceMode, int numWorkers) {
+        super(inferenceAdapter, serializer, deserializer, binarySerializer, binaryDeserializer, port);
 
         this.mlnModel = mlnModel;
         this.inferenceMode = inferenceMode;
         this.numWorkers = numWorkers;
     }
 
-    protected JsonModelServer(@NonNull ParallelInference pi, InferenceAdapter<I, O> inferenceAdapter, JsonSerializer<O> serializer, JsonDeserializer<I> deserializer, int port) {
+    protected JsonModelServer(@NonNull ParallelInference pi, InferenceAdapter<I, O> inferenceAdapter,
-        super(inferenceAdapter, serializer, deserializer, port);
+                              JsonSerializer<O> serializer, JsonDeserializer<I> deserializer,
+                              BinarySerializer<O> binarySerializer, BinaryDeserializer<I> binaryDeserializer,
+                              int port) {
+        super(inferenceAdapter, serializer, deserializer, binarySerializer, binaryDeserializer, port);
 
         this.parallelInference = pi;
     }
@@ -139,6 +153,8 @@ public class JsonModelServer<I, O> extends SameDiffJsonModelServer<I, O> {
                         .parallelEnabled(true)
                         .serializer(serializer)
                         .deserializer(deserializer)
+                        .binarySerializer(binarySerializer)
+                        .binaryDeserializer(binaryDeserializer)
                         .inferenceAdapter(inferenceAdapter)
                         .build();
         }
@@ -147,6 +163,8 @@ public class JsonModelServer<I, O> extends SameDiffJsonModelServer<I, O> {
                         .parallelEnabled(false)
                         .serializer(serializer)
                         .deserializer(deserializer)
+                        .binarySerializer(binarySerializer)
+                        .binaryDeserializer(binaryDeserializer)
                         .inferenceAdapter(inferenceAdapter)
                         .build();
         }
@@ -175,6 +193,8 @@ public class JsonModelServer<I, O> extends SameDiffJsonModelServer<I, O> {
         private InferenceAdapter<I, O> inferenceAdapter;
         private JsonSerializer<O> serializer;
         private JsonDeserializer<I> deserializer;
+        private BinarySerializer<O> binarySerializer;
+        private BinaryDeserializer<I> binaryDeserializer;
 
         private InputAdapter<I> inputAdapter;
         private OutputAdapter<O> outputAdapter;
@@ -238,7 +258,9 @@ public class JsonModelServer<I, O> extends SameDiffJsonModelServer<I, O> {
         }
 
         /**
-         * This method allows you to specify serializer
+         * This method allows you to specify JSON serializer.
+         * Incompatible with {@link #outputBinarySerializer(BinarySerializer)}
+         * Only one serializer - deserializer pair can be used by client and server.
          *
          * @param serializer
          * @return
@@ -249,7 +271,9 @@ public class JsonModelServer<I, O> extends SameDiffJsonModelServer<I, O> {
         }
 
         /**
-         * This method allows you to specify deserializer
+         * This method allows you to specify JSON deserializer.
+         * Incompatible with {@link #inputBinaryDeserializer(BinaryDeserializer)}
+         * Only one serializer - deserializer pair can be used by client and server.
          *
          * @param deserializer
          * @return
@@ -259,6 +283,32 @@ public class JsonModelServer<I, O> extends SameDiffJsonModelServer<I, O> {
             return this;
         }
 
+        /**
+         * This method allows you to specify binary serializer.
+         * Incompatible with {@link #outputSerializer(JsonSerializer)}
+         * Only one serializer - deserializer pair can be used by client and server.
+         *
+         * @param serializer
+         * @return
+         */
+        public Builder<I,O> outputBinarySerializer(@NonNull BinarySerializer<O> serializer) {
+            this.binarySerializer = serializer;
+            return this;
+        }
+
+        /**
+         * This method allows you to specify binary deserializer
+         * Incompatible with {@link #inputDeserializer(JsonDeserializer)}
+         * Only one serializer - deserializer pair can be used by client and server.
+         *
+         * @param deserializer
+         * @return
+         */
+        public Builder<I,O> inputBinaryDeserializer(@NonNull BinaryDeserializer<I> deserializer) {
+            this.binaryDeserializer = deserializer;
+            return this;
+        }
+
         /**
          * This method allows you to specify inference mode for parallel mode. See {@link InferenceMode} for more details
          *
@@ -375,17 +425,24 @@ public class JsonModelServer<I, O> extends SameDiffJsonModelServer<I, O> {
                     throw new IllegalArgumentException("Either InferenceAdapter<I,O> or InputAdapter<I> + OutputAdapter<O> should be configured");
             }
 
+            JsonModelServer server = null;
             if (sdModel != null) {
-                Preconditions.checkArgument(orderedOutputNodes != null && orderedOutputNodes.length > 0, "For SameDiff model serving OutputNodes should be defined");
+                server = new JsonModelServer<I, O>(sdModel, inferenceAdapter, serializer, deserializer, binarySerializer, binaryDeserializer, port, orderedInputNodes, orderedOutputNodes);
-                return new JsonModelServer<I, O>(sdModel, inferenceAdapter, serializer, deserializer, port, orderedInputNodes, orderedOutputNodes);
+            }
-            } else if (cgModel != null)
+            else if (cgModel != null) {
-                return new JsonModelServer<I,O>(cgModel, inferenceAdapter, serializer, deserializer, port, inferenceMode, numWorkers);
+                server = new JsonModelServer<I, O>(cgModel, inferenceAdapter, serializer, deserializer, binarySerializer, binaryDeserializer, port, inferenceMode, numWorkers);
-            else if (mlnModel != null)
+            }
-                return new JsonModelServer<I,O>(mlnModel, inferenceAdapter, serializer, deserializer, port, inferenceMode, numWorkers);
+            else if (mlnModel != null) {
-            else if (pi != null)
+                server = new JsonModelServer<I, O>(mlnModel, inferenceAdapter, serializer, deserializer, binarySerializer, binaryDeserializer, port, inferenceMode, numWorkers);
-                return new JsonModelServer<I,O>(pi, inferenceAdapter, serializer, deserializer, port);
+            }
+            else if (pi != null) {
+                 server = new JsonModelServer<I, O>(pi, inferenceAdapter, serializer, deserializer, binarySerializer, binaryDeserializer, port);
+            }
               else
                  throw new IllegalStateException("No models were defined for JsonModelServer");
+
+            server.enabledParallel = parallelMode;
+            return server;
         }
     }
 

deeplearning4j/deeplearning4j-remote/deeplearning4j-json-server/src/test/java/org/deeplearning4j/remote/BinaryModelServerTest.java

@@ -0,0 +1,276 @@
+package org.deeplearning4j.remote;
+
+import lombok.val;
+import org.datavec.image.loader.Java2DNativeImageLoader;
+import org.deeplearning4j.nn.graph.ComputationGraph;
+import org.deeplearning4j.nn.multilayer.MultiLayerNetwork;
+import org.deeplearning4j.remote.helpers.ImageConversionUtils;
+import org.deeplearning4j.util.ModelSerializer;
+import org.junit.After;
+import org.junit.Test;
+import org.nd4j.adapters.InferenceAdapter;
+import org.nd4j.linalg.api.ndarray.INDArray;
+import org.nd4j.linalg.dataset.MultiDataSet;
+import org.nd4j.linalg.io.ClassPathResource;
+import org.nd4j.remote.clients.JsonRemoteInference;
+import org.nd4j.remote.clients.serde.BinaryDeserializer;
+import org.nd4j.remote.clients.serde.BinarySerializer;
+import org.nd4j.remote.clients.serde.JsonDeserializer;
+import org.nd4j.remote.clients.serde.JsonSerializer;
+import org.nd4j.remote.clients.serde.impl.IntegerSerde;
+import org.nd4j.shade.jackson.databind.ObjectMapper;
+
+import javax.imageio.ImageIO;
+import java.awt.image.BufferedImage;
+import java.io.*;
+import java.nio.Buffer;
+import java.util.concurrent.Future;
+import java.util.concurrent.TimeUnit;
+
+import static org.deeplearning4j.parallelism.inference.InferenceMode.SEQUENTIAL;
+import static org.junit.Assert.*;
+
+public class BinaryModelServerTest {
+    private final int PORT = 18080;
+
+    @After
+    public void pause() throws Exception {
+        // TODO: the same port was used in previous test and not accessible immediately. Might be better solution.
+        TimeUnit.SECONDS.sleep(2);
+    }
+
+    // Internal test for locally defined serializers
+    @Test
+    public void testBufferedImageSerde() {
+            BinarySerializer<BufferedImage> serde = new BinaryModelServerTest.BufferedImageSerde();
+            BufferedImage image = ImageConversionUtils.makeRandomBufferedImage(28,28,1);
+            byte[] serialized = serde.serialize(image);
+
+            BufferedImage deserialized = ((BufferedImageSerde) serde).deserialize(serialized);
+            int originalSize = image.getData().getDataBuffer().getSize();
+            assertEquals(originalSize, deserialized.getData().getDataBuffer().getSize());
+            for (int i = 0; i < originalSize; ++i) {
+                assertEquals(deserialized.getData().getDataBuffer().getElem(i),
+                             image.getData().getDataBuffer().getElem(i));
+            }
+    }
+
+    @Test
+    public void testImageToINDArray() {
+        INDArray data = ImageConversionUtils.makeRandomImageAsINDArray(28,28,1);
+        assertNotNull(data);
+    }
+
+    @Test
+    public void testMlnMnist_ImageInput() throws Exception {
+
+        val modelFile = new ClassPathResource("models/mnist/mnist-model.zip").getFile();
+        MultiLayerNetwork net = ModelSerializer.restoreMultiLayerNetwork(modelFile);
+
+        val server = new JsonModelServer.Builder<BufferedImage, Integer>(net)
+                .outputSerializer(new IntegerSerde())
+                .inputBinaryDeserializer(new BufferedImageSerde())
+                .inferenceAdapter(new InferenceAdapter<BufferedImage, Integer>() {
+                    @Override
+                    public MultiDataSet apply(BufferedImage input) {
+                        INDArray data = null;
+                        try {
+                            data = new Java2DNativeImageLoader().asMatrix(input);
+                            data = data.reshape(1, 784);
+                        }
+                        catch (IOException e) {
+                            throw new RuntimeException(e);
+                        }
+                        return new MultiDataSet(data, null);
+                    }
+
+                    @Override
+                    public Integer apply(INDArray... nnOutput) {
+                        return nnOutput[0].argMax().getInt(0);
+                    }
+                })
+                .port(PORT)
+                .inferenceMode(SEQUENTIAL)
+                .numWorkers(1)
+                .parallelMode(false)
+                .build();
+
+        val client = JsonRemoteInference.<BufferedImage, Integer>builder()
+                .endpointAddress("http://localhost:" + PORT + "/v1/serving")
+                .inputBinarySerializer(new BufferedImageSerde())
+                .outputDeserializer(new IntegerSerde())
+                .build();
+
+        try {
+            server.start();
+            BufferedImage image = ImageConversionUtils.makeRandomBufferedImage(28,28,1);
+            Integer result = client.predict(image);
+            assertNotNull(result);
+
+            File file = new ClassPathResource("datavec-local/imagetest/0/b.bmp").getFile();
+            image = ImageIO.read(new FileInputStream(file));
+            result = client.predict(image);
+            assertEquals(new Integer(0), result);
+
+            file = new ClassPathResource("datavec-local/imagetest/1/a.bmp").getFile();
+            image = ImageIO.read(new FileInputStream(file));
+            result = client.predict(image);
+            assertEquals(new Integer(1), result);
+
+        } catch (Exception e){
+            e.printStackTrace();
+            throw e;
+        } finally {
+            server.stop();
+        }
+    }
+
+    @Test
+    public void testMlnMnist_ImageInput_Async() throws Exception {
+
+        val modelFile = new ClassPathResource("models/mnist/mnist-model.zip").getFile();
+        MultiLayerNetwork net = ModelSerializer.restoreMultiLayerNetwork(modelFile);
+
+        val server = new JsonModelServer.Builder<BufferedImage, Integer>(net)
+                .outputSerializer(new IntegerSerde())
+                .inputBinaryDeserializer(new BufferedImageSerde())
+                .inferenceAdapter(new InferenceAdapter<BufferedImage, Integer>() {
+                    @Override
+                    public MultiDataSet apply(BufferedImage input) {
+                        INDArray data = null;
+                        try {
+                            data = new Java2DNativeImageLoader().asMatrix(input);
+                            data = data.reshape(1, 784);
+                        }
+                        catch (IOException e) {
+                            throw new RuntimeException(e);
+                        }
+                        return new MultiDataSet(data, null);
+                    }
+
+                    @Override
+                    public Integer apply(INDArray... nnOutput) {
+                        return nnOutput[0].argMax().getInt(0);
+                    }
+                })
+                .port(PORT)
+                .inferenceMode(SEQUENTIAL)
+                .numWorkers(1)
+                .parallelMode(false)
+                .build();
+
+        val client = JsonRemoteInference.<BufferedImage, Integer>builder()
+                .endpointAddress("http://localhost:" + PORT + "/v1/serving")
+                .inputBinarySerializer(new BufferedImageSerde())
+                .outputDeserializer(new IntegerSerde())
+                .build();
+
+        try {
+            server.start();
+            BufferedImage[] images = new BufferedImage[3];
+            images[0] = ImageConversionUtils.makeRandomBufferedImage(28,28,1);
+
+            File file = new ClassPathResource("datavec-local/imagetest/0/b.bmp").getFile();
+            images[1] = ImageIO.read(new FileInputStream(file));
+
+            file = new ClassPathResource("datavec-local/imagetest/1/a.bmp").getFile();
+            images[2] = ImageIO.read(new FileInputStream(file));
+
+            Future<Integer>[] results = new Future[3];
+            for (int i = 0; i < images.length; ++i) {
+                results[i] = client.predictAsync(images[i]);
+                assertNotNull(results[i]);
+            }
+
+            assertNotNull(results[0].get());
+            assertEquals(new Integer(0), results[1].get());
+            assertEquals(new Integer(1), results[2].get());
+
+        } catch (Exception e){
+            e.printStackTrace();
+            throw e;
+        } finally {
+            server.stop();
+        }
+    }
+
+    @Test
+    public void testBinaryIn_BinaryOut() throws Exception {
+
+        val modelFile = new ClassPathResource("models/mnist/mnist-model.zip").getFile();
+        MultiLayerNetwork net = ModelSerializer.restoreMultiLayerNetwork(modelFile);
+
+        val server = new JsonModelServer.Builder<BufferedImage, BufferedImage>(net)
+                .outputBinarySerializer(new BufferedImageSerde())
+                .inputBinaryDeserializer(new BufferedImageSerde())
+                .inferenceAdapter(new InferenceAdapter<BufferedImage, BufferedImage>() {
+                    @Override
+                    public MultiDataSet apply(BufferedImage input) {
+                        INDArray data = null;
+                        try {
+                            data = new Java2DNativeImageLoader().asMatrix(input);
+                        }
+                        catch (IOException e) {
+                            throw new RuntimeException(e);
+                        }
+                        return new MultiDataSet(data, null);
+                    }
+
+                    @Override
+                    public BufferedImage apply(INDArray... nnOutput) {
+                        return ImageConversionUtils.makeRandomBufferedImage(28,28,3);
+                    }
+                })
+                .port(PORT)
+                .inferenceMode(SEQUENTIAL)
+                .numWorkers(1)
+                .parallelMode(false)
+                .build();
+
+        val client = JsonRemoteInference.<BufferedImage, BufferedImage>builder()
+                .endpointAddress("http://localhost:" + PORT + "/v1/serving")
+                .inputBinarySerializer(new BufferedImageSerde())
+                .outputBinaryDeserializer(new BufferedImageSerde())
+                .build();
+
+        try {
+            server.start();
+            BufferedImage image = ImageConversionUtils.makeRandomBufferedImage(28,28,1);
+            BufferedImage result = client.predict(image);
+            assertNotNull(result);
+            assertEquals(28, result.getHeight());
+            assertEquals(28, result.getWidth());
+
+        } catch (Exception e){
+            e.printStackTrace();
+            throw e;
+        } finally {
+            server.stop();
+        }
+    }
+
+    private static class BufferedImageSerde implements BinarySerializer<BufferedImage>, BinaryDeserializer<BufferedImage> {
+
+        @Override
+        public BufferedImage deserialize(byte[] buffer) {
+            try {
+                BufferedImage img = ImageIO.read(new ByteArrayInputStream(buffer));
+                return img;
+            } catch (IOException e){
+                throw new RuntimeException(e);
+            }
+        }
+
+        @Override
+        public byte[] serialize(BufferedImage image) {
+            try{
+                val baos = new ByteArrayOutputStream();
+                ImageIO.write(image, "bmp", baos);
+                byte[] bytes = baos.toByteArray();
+                return bytes;
+            } catch (IOException e){
+                throw new RuntimeException(e);
+            }
+        }
+    }
+}

deeplearning4j/deeplearning4j-remote/deeplearning4j-json-server/src/test/java/org/deeplearning4j/remote/helpers/ImageConversionUtils.java

@@ -0,0 +1,82 @@
+package org.deeplearning4j.remote.helpers;
+
+import lombok.val;
+import org.bytedeco.javacpp.indexer.UByteIndexer;
+import org.bytedeco.javacv.Java2DFrameConverter;
+import org.bytedeco.javacv.OpenCVFrameConverter;
+import org.bytedeco.opencv.opencv_core.Mat;
+import org.datavec.image.loader.Java2DNativeImageLoader;
+import org.datavec.image.loader.NativeImageLoader;
+import org.nd4j.linalg.api.ndarray.INDArray;
+
+import java.awt.image.BufferedImage;
+import java.io.IOException;
+import java.util.Random;
+
+import static org.bytedeco.opencv.global.opencv_core.CV_8UC;
+
+public class ImageConversionUtils {
+
+    public static Mat makeRandomImage(int height, int width, int channels) {
+        if (height <= 0) {
+
+            height = new Random().nextInt() % 100 + 100;
+        }
+        if (width <= 0) {
+            width = new Random().nextInt() % 100 + 100;
+        }
+
+        Mat img = new Mat(height, width, CV_8UC(channels));
+        UByteIndexer idx = img.createIndexer();
+        for (int i = 0; i < height; i++) {
+            for (int j = 0; j < width; j++) {
+                for (int k = 0; k < channels; k++) {
+                    idx.put(i, j, k, new Random().nextInt());
+                }
+            }
+        }
+        return img;
+    }
+
+    public static BufferedImage makeRandomBufferedImage(int height, int width, int channels) {
+        Mat img = makeRandomImage(height, width, channels);
+
+        OpenCVFrameConverter.ToMat c = new OpenCVFrameConverter.ToMat();
+        Java2DFrameConverter c2 = new Java2DFrameConverter();
+
+        return c2.convert(c.convert(img));
+    }
+
+    public static INDArray convert(BufferedImage image) {
+        INDArray retVal = null;
+        try {
+            retVal = new Java2DNativeImageLoader(image.getHeight(), image.getWidth(), image.getRaster().getNumBands()).
+                    asRowVector(image);
+        }
+        catch (IOException e) {
+            throw new RuntimeException(e);
+        }
+        return retVal;
+    }
+
+    public static INDArray convert(Mat image) {
+        INDArray retVal = null;
+        try {
+            new NativeImageLoader().asRowVector(image);
+        }
+        catch (IOException e) {
+            throw new RuntimeException(e);
+        }
+        return retVal;
+    }
+
+    public static BufferedImage convert(INDArray input) {
+        return new Java2DNativeImageLoader(input.rows(),input.columns()).asBufferedImage(input);
+    }
+
+    public static INDArray makeRandomImageAsINDArray(int height, int width, int channels) {
+        val image = makeRandomBufferedImage(height, width, channels);
+        INDArray retVal = convert(image);
+        return retVal;
+    }
+}

deeplearning4j/pom.xml

@@ -69,7 +69,7 @@
         </developer>
         <developer>
             <id>raver119</id>
-            <name>raver119</name>
+            <name>Vyacheslav Kokorin</name>
         </developer>
         <developer>
             <id>saudet</id>

nd4j/nd4j-remote/nd4j-json-client/src/main/java/org/nd4j/remote/clients/JsonRemoteInference.java

@@ -24,10 +24,12 @@ import lombok.NonNull;
 import lombok.extern.slf4j.Slf4j;
 import lombok.val;
 import org.json.JSONObject;
-import org.nd4j.remote.clients.serde.JsonDeserializer;
+import org.nd4j.remote.clients.serde.*;
-import org.nd4j.remote.clients.serde.JsonSerializer;
 
 import java.io.IOException;
+import java.io.InputStream;
+import java.util.Collections;
+import java.util.List;
 import java.util.concurrent.ExecutionException;
 import java.util.concurrent.Future;
 import java.util.concurrent.TimeUnit;
@@ -49,21 +51,57 @@ import java.util.concurrent.TimeoutException;
 @Slf4j
 public class JsonRemoteInference<I, O> {
     private String endpointAddress;
+    // JSON serializer/deserializer and binary serializer/deserializer are mutually exclusive.
     private JsonSerializer<I> serializer;
     private JsonDeserializer<O> deserializer;
+    private BinarySerializer<I> binarySerializer;
+    private BinaryDeserializer<O> binaryDeserializer;
+
+    private final static String APPLICATION_JSON = "application/json";
+    private final static String APPLICATION_OCTET_STREAM = "application/octet-stream";
 
     @Builder
-    public JsonRemoteInference(@NonNull String endpointAddress, @NonNull JsonSerializer<I> inputSerializer, @NonNull JsonDeserializer<O> outputDeserializer) {
+    public JsonRemoteInference(@NonNull String endpointAddress,
+                               JsonSerializer<I> inputSerializer, JsonDeserializer<O> outputDeserializer,
+                               BinarySerializer<I> inputBinarySerializer, BinaryDeserializer<O> outputBinaryDeserializer) {
+
         this.endpointAddress = endpointAddress;
         this.serializer = inputSerializer;
         this.deserializer = outputDeserializer;
+        this.binarySerializer = inputBinarySerializer;
+        this.binaryDeserializer = outputBinaryDeserializer;
+
+        if (serializer != null && binarySerializer != null || serializer == null && binarySerializer == null)
+            throw new IllegalStateException("Binary and JSON serializers/deserializers are mutually exclusive and mandatory.");
     }
 
+
     private O processResponse(HttpResponse<String> response) throws IOException {
         if (response.getStatus() != 200)
             throw new IOException("Inference request returned bad error code: " + response.getStatus());
 
         O result = deserializer.deserialize(response.getBody());
+
+        if (result == null) {
+            throw new IOException("Deserialization failed!");
+        }
+        return result;
+    }
+
+    private O processResponseBinary(HttpResponse<InputStream> response) throws IOException {
+        if (response.getStatus() != 200)
+            throw new IOException("Inference request returned bad error code: " + response.getStatus());
+
+        List<String> values = response.getHeaders().get("Content-Length");
+        if (values == null || values.size() < 1) {
+            throw new IOException("Content-Length is required for binary data");
+        }
+
+        String strLength = values.get(0);
+        byte[] bytes = new byte[Integer.parseInt(strLength)];
+        response.getBody().read(bytes);
+        O result = binaryDeserializer.deserialize(bytes);
+
         if (result == null) {
             throw new IOException("Deserialization failed!");
         }
@@ -79,12 +117,30 @@ public class JsonRemoteInference<I, O> {
      */
     public O predict(I input) throws IOException {
         try {
-            val stringResult = Unirest.post(endpointAddress)
+            if (binarySerializer != null && binaryDeserializer != null) {
-                    .header("Content-Type", "application/json")
+                HttpResponse<InputStream> response =
-                    .header("Accept", "application/json")
+                        Unirest.post(endpointAddress)
+                                .header("Content-Type", APPLICATION_OCTET_STREAM)
+                                .header("Accept", APPLICATION_OCTET_STREAM)
+                                .body(binarySerializer.serialize(input)).asBinary();
+                return processResponseBinary(response);
+            }
+            else if (binarySerializer != null && binaryDeserializer == null) {
+                HttpResponse<String> response =
+                        Unirest.post(endpointAddress)
+                                .header("Content-Type", APPLICATION_OCTET_STREAM)
+                                .header("Accept", APPLICATION_OCTET_STREAM)
+                                .body(binarySerializer.serialize(input)).asString();
+                return processResponse(response);
+            }
+            else {
+                HttpResponse<String> response = Unirest.post(endpointAddress)
+                                .header("Content-Type", APPLICATION_JSON)
+                                .header("Accept", APPLICATION_JSON)
                                 .body(new JSONObject(serializer.serialize(input))).asString();
+                return processResponse(response);
+            }
 
-            return processResponse(stringResult);
         } catch (UnirestException e) {
             throw new IOException(e);
         }
@@ -96,11 +152,19 @@ public class JsonRemoteInference<I, O> {
      * @return
      */
     public Future<O> predictAsync(I input) {
-        val stringResult = Unirest.post(endpointAddress)
+
+        Future<HttpResponse<String>> response = binarySerializer != null ?
+                Unirest.post(endpointAddress)
+                .header("Content-Type", "application/octet-stream")
+                .header("Accept", "application/octet-stream")
+                .body(binarySerializer.serialize(input)).asStringAsync() :
+
+                Unirest.post(endpointAddress)
                 .header("Content-Type", "application/json")
                 .header("Accept", "application/json")
                 .body(new JSONObject(serializer.serialize(input))).asStringAsync();
-        return new InferenceFuture(stringResult);
+
+        return new InferenceFuture(response);
     }
 
     /**
@@ -151,3 +215,4 @@ public class JsonRemoteInference<I, O> {
         }
     }
 }
+

nd4j/nd4j-remote/nd4j-json-client/src/main/java/org/nd4j/remote/clients/serde/BinaryDeserializer.java

@@ -0,0 +1,32 @@
+/*******************************************************************************
+ * Copyright (c) 2015-2018 Skymind, Inc.
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Apache License, Version 2.0 which is available at
+ * https://www.apache.org/licenses/LICENSE-2.0.
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+ * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+ * License for the specific language governing permissions and limitations
+ * under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ ******************************************************************************/
+package org.nd4j.remote.clients.serde;
+
+/**
+ * This interface describes basic binary deserializer interface used for remote inference
+ * @param <T> type of the deserializable class
+ *
+ * @author Alexander Stoyakin
+ */
+public interface BinaryDeserializer<T> {
+
+        /**
+         * This method deserializes binary data to arbitrary object.
+         * @param byte buffer
+         * @return deserialized object
+         */
+        T deserialize(byte[] buffer);
+}

nd4j/nd4j-remote/nd4j-json-client/src/main/java/org/nd4j/remote/clients/serde/BinarySerializer.java

@@ -0,0 +1,34 @@
+/*******************************************************************************
+ * Copyright (c) 2015-2018 Skymind, Inc.
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Apache License, Version 2.0 which is available at
+ * https://www.apache.org/licenses/LICENSE-2.0.
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+ * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+ * License for the specific language governing permissions and limitations
+ * under the License.
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ ******************************************************************************/
+
+package org.nd4j.remote.clients.serde;
+
+/**
+ * This interface describes basic binary serializer interface used for remote inference
+ * @param <T> type of the serializable class
+ *
+ * @author Alexander Stoyakin
+ */
+public interface BinarySerializer<T> {
+
+    /**
+     * This method serializes given object into byte buffer
+     *
+     * @param o object to be serialized
+     * @return
+     */
+    byte[] serialize(T o);
+}

nd4j/nd4j-remote/nd4j-json-server/src/main/java/org/nd4j/remote/SameDiffJsonModelServer.java

@@ -26,6 +26,8 @@ import org.nd4j.autodiff.samediff.SameDiff;
 import org.nd4j.base.Preconditions;
 import org.nd4j.linalg.api.ndarray.INDArray;
 import org.nd4j.linalg.dataset.MultiDataSet;
+import org.nd4j.remote.clients.serde.BinaryDeserializer;
+import org.nd4j.remote.clients.serde.BinarySerializer;
 import org.nd4j.remote.clients.serde.JsonDeserializer;
 import org.nd4j.remote.clients.serde.JsonSerializer;
 import org.nd4j.adapters.InferenceAdapter;
@@ -51,6 +53,8 @@ public class SameDiffJsonModelServer<I, O> {
     protected SameDiff sdModel;
     protected final JsonSerializer<O> serializer;
     protected final JsonDeserializer<I> deserializer;
+    protected final BinarySerializer<O> binarySerializer;
+    protected final BinaryDeserializer<I> binaryDeserializer;
     protected final InferenceAdapter<I, O> inferenceAdapter;
     protected final int port;
 
@@ -64,9 +68,18 @@ public class SameDiffJsonModelServer<I, O> {
     protected String[] orderedInputNodes;
     protected String[] orderedOutputNodes;
 
-    protected SameDiffJsonModelServer(@NonNull InferenceAdapter<I, O> inferenceAdapter, @NonNull JsonSerializer<O> serializer, @NonNull JsonDeserializer<I> deserializer, int port) {
+    protected SameDiffJsonModelServer(@NonNull InferenceAdapter<I, O> inferenceAdapter,
+                                      JsonSerializer<O> serializer, JsonDeserializer<I> deserializer,
+                                      BinarySerializer<O> binarySerializer, BinaryDeserializer<I> binaryDeserializer,
+                                      int port) {
         Preconditions.checkArgument(port > 0 && port < 65535, "TCP port must be in range of 0..65535");
+        Preconditions.checkArgument(serializer == null && binarySerializer == null ||
+                                        serializer != null && binarySerializer == null ||
+                                        serializer == null && binarySerializer != null,
+                                "JSON and binary serializers/deserializers are mutually exclusive and mandatory.");
 
+        this.binarySerializer = binarySerializer;
+        this.binaryDeserializer = binaryDeserializer;
         this.inferenceAdapter = inferenceAdapter;
         this.serializer = serializer;
         this.deserializer = deserializer;
@@ -74,8 +87,11 @@ public class SameDiffJsonModelServer<I, O> {
     }
 
     //@Builder
-    public SameDiffJsonModelServer(SameDiff sdModel, @NonNull InferenceAdapter<I, O> inferenceAdapter, @NonNull JsonSerializer<O> serializer, @NonNull JsonDeserializer<I> deserializer, int port, String[] orderedInputNodes, @NonNull String[] orderedOutputNodes) {
+    public SameDiffJsonModelServer(SameDiff sdModel, @NonNull InferenceAdapter<I, O> inferenceAdapter,
-        this(inferenceAdapter, serializer, deserializer, port);
+                                   JsonSerializer<O> serializer, JsonDeserializer<I> deserializer,
+                                   BinarySerializer<O> binarySerializer, BinaryDeserializer<I> binaryDeserializer,
+                                   int port, String[] orderedInputNodes, @NonNull String[] orderedOutputNodes) {
+        this(inferenceAdapter, serializer, deserializer, binarySerializer, binaryDeserializer, port);
         this.sdModel = sdModel;
         this.orderedInputNodes = orderedInputNodes;
         this.orderedOutputNodes = orderedOutputNodes;
@@ -282,7 +298,7 @@ public class SameDiffJsonModelServer<I, O> {
                 } else
                     throw new IllegalArgumentException("Either InferenceAdapter<I,O> or InputAdapter<I> + OutputAdapter<O> should be configured");
             }
-            return new SameDiffJsonModelServer<I,O>(sameDiff, inferenceAdapter, serializer, deserializer, port, orderedInputNodes, orderedOutputNodes);
+            return new SameDiffJsonModelServer<I,O>(sameDiff, inferenceAdapter, serializer, deserializer, null, null, port, orderedInputNodes, orderedOutputNodes);
         }
     }
 }

nd4j/nd4j-remote/nd4j-json-server/src/main/java/org/nd4j/remote/serving/SameDiffServlet.java

@@ -21,6 +21,8 @@ import lombok.extern.slf4j.Slf4j;
 import org.apache.commons.lang3.StringUtils;
 import org.nd4j.autodiff.samediff.SameDiff;
 import org.nd4j.linalg.api.ndarray.INDArray;
+import org.nd4j.remote.clients.serde.BinaryDeserializer;
+import org.nd4j.remote.clients.serde.BinarySerializer;
 import org.nd4j.remote.clients.serde.JsonDeserializer;
 import org.nd4j.remote.clients.serde.JsonSerializer;
 import org.nd4j.adapters.InferenceAdapter;
@@ -35,6 +37,7 @@ import java.io.InputStreamReader;
 import java.util.LinkedHashMap;
 
 import static javax.ws.rs.core.MediaType.APPLICATION_JSON;
+import static javax.ws.rs.core.MediaType.APPLICATION_OCTET_STREAM;
 
 /**
  * This servlet provides SameDiff model serving capabilities
@@ -50,9 +53,14 @@ import static javax.ws.rs.core.MediaType.APPLICATION_JSON;
 @Builder
 public class SameDiffServlet<I, O> implements ModelServingServlet<I, O> {
 
+    protected static final String typeJson = APPLICATION_JSON;
+    protected static final String typeBinary = APPLICATION_OCTET_STREAM;
+
     protected SameDiff sdModel;
     protected JsonSerializer<O> serializer;
     protected JsonDeserializer<I> deserializer;
+    protected BinarySerializer<O> binarySerializer;
+    protected BinaryDeserializer<I> binaryDeserializer;
     protected InferenceAdapter<I, O> inferenceAdapter;
 
     protected String[] orderedInputNodes;
@@ -60,14 +68,36 @@ public class SameDiffServlet<I, O> implements ModelServingServlet<I, O> {
 
     protected final static String SERVING_ENDPOINT = "/v1/serving";
     protected final static String LISTING_ENDPOINT = "/v1";
-    protected final static int PAYLOAD_SIZE_LIMIT = 10 * 1024; // TODO: should be customizable
+    protected final static int PAYLOAD_SIZE_LIMIT = 10 * 1024 * 1024; // TODO: should be customizable
 
-    protected SameDiffServlet(@NonNull InferenceAdapter<I, O> inferenceAdapter, @NonNull JsonSerializer<O> serializer, @NonNull JsonDeserializer<I> deserializer){
+    protected SameDiffServlet(@NonNull InferenceAdapter<I, O> inferenceAdapter, JsonSerializer<O> serializer, JsonDeserializer<I> deserializer){
         this.serializer = serializer;
         this.deserializer = deserializer;
         this.inferenceAdapter = inferenceAdapter;
     }
 
+    protected SameDiffServlet(@NonNull InferenceAdapter<I, O> inferenceAdapter,
+                              BinarySerializer<O> serializer, BinaryDeserializer<I> deserializer){
+        this.binarySerializer = serializer;
+        this.binaryDeserializer = deserializer;
+        this.inferenceAdapter = inferenceAdapter;
+    }
+
+    protected SameDiffServlet(@NonNull InferenceAdapter<I, O> inferenceAdapter,
+                              JsonSerializer<O> jsonSerializer, JsonDeserializer<I> jsonDeserializer,
+                              BinarySerializer<O> binarySerializer, BinaryDeserializer<I> binaryDeserializer){
+
+        this.serializer = jsonSerializer;
+        this.deserializer = jsonDeserializer;
+        this.binarySerializer = binarySerializer;
+        this.binaryDeserializer = binaryDeserializer;
+        this.inferenceAdapter = inferenceAdapter;
+
+        if (serializer != null && binarySerializer != null || serializer == null && binarySerializer == null)
+            throw new IllegalStateException("Binary and JSON serializers/deserializers are mutually exclusive and mandatory.");
+    }
+
+
     @Override
     public void init(ServletConfig servletConfig) throws ServletException {
         //
@@ -108,7 +138,7 @@ public class SameDiffServlet<I, O> implements ModelServingServlet<I, O> {
     protected boolean validateRequest(HttpServletRequest request, HttpServletResponse response)
                                                                             throws IOException{
         val contentType = request.getContentType();
-        if (!StringUtils.equals(contentType, APPLICATION_JSON)) {
+        if (!StringUtils.equals(contentType, typeJson)) {
             sendBadContentType(contentType, response);
             int contentLength = request.getContentLength();
             if (contentLength > PAYLOAD_SIZE_LIMIT) {
@@ -125,7 +155,7 @@ public class SameDiffServlet<I, O> implements ModelServingServlet<I, O> {
         String path = request.getPathInfo();
         if (path.equals(LISTING_ENDPOINT)) {
             val contentType = request.getContentType();
-            if (!StringUtils.equals(contentType, APPLICATION_JSON)) {
+            if (!StringUtils.equals(contentType, typeJson)) {
                 sendBadContentType(contentType, response);
             }
             processorReturned = processor.listEndpoints();
@@ -147,7 +177,7 @@ public class SameDiffServlet<I, O> implements ModelServingServlet<I, O> {
         String path = request.getPathInfo();
         if (path.equals(SERVING_ENDPOINT)) {
             val contentType = request.getContentType();
-            /*Preconditions.checkArgument(StringUtils.equals(contentType, APPLICATION_JSON),
+            /*Preconditions.checkArgument(StringUtils.equals(contentType, typeJson),
                     "Content type is " + contentType);*/
             if (validateRequest(request,response)) {
                 val stream = request.getInputStream();