You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@mxnet.apache.org by GitBox <gi...@apache.org> on 2018/11/12 19:58:06 UTC
[GitHub] aaronmarkham commented on a change in pull request #13201:
[MXNET-1187] Added Java SSD Inference Tutorial for website
aaronmarkham commented on a change in pull request #13201: [MXNET-1187] Added Java SSD Inference Tutorial for website
URL: https://github.com/apache/incubator-mxnet/pull/13201#discussion_r232792073
##########
File path: docs/tutorials/java/ssd_inference.md
##########
@@ -0,0 +1,182 @@
+# Multi Object Detection using pre-trained SSD Model via Java Inference APIs
+
+This tutorial shows how to use MXNet Java Inference APIs to run inference on a pre-trained Single Shot Detection (SSD) Model.
+
+The SSD model is trained on the Pascal VOC 2012 dataset. The network is a SSD model built on Resnet50 as base network to extract image features. The model is trained to detect the following entities (classes): ['aeroplane', 'bicycle', 'bird', 'boat', 'bottle', 'bus', 'car', 'cat', 'chair', 'cow', 'diningtable', 'dog', 'horse', 'motorbike', 'person', 'pottedplant', 'sheep', 'sofa', 'train', 'tvmonitor']. For more details about the model, you can refer to the [MXNet SSD example](https://github.com/apache/incubator-mxnet/tree/master/example/ssd).
+
+## Pre-Requisites
+
+To complete this tutorial, we need :
+* [MXNet Java Setup on IntelliJ IDEA](/java/mxnet_java_on_intellij.html)
+* [wget](https://www.gnu.org/software/wget/) To download model artifacts
+* SSD Model artifacts
+ * Use the following script to get the SSD Model files :
+ ```bash
+ data_path=/tmp/resnet50_ssd
+ mkdir -p "$data_path"
+ wget https://s3.amazonaws.com/model-server/models/resnet50_ssd/resnet50_ssd_model-symbol.json -P $data_path
+ wget https://s3.amazonaws.com/model-server/models/resnet50_ssd/resnet50_ssd_model-0000.params -P $data_path
+ wget https://s3.amazonaws.com/model-server/models/resnet50_ssd/synset.txt -P $data_path
+ ```
+* Dataset : A few sample images to run inference on.
+ * Use the following script to download sample images :
+ ```bash
+ image_path=/tmp/resnet50_ssd/images
+ mkdir -p "$image_path"
+ cd $image_path
+ wget https://cloud.githubusercontent.com/assets/3307514/20012567/cbb60336-a27d-11e6-93ff-cbc3f09f5c9e.jpg -O dog.jpg
+ wget https://cloud.githubusercontent.com/assets/3307514/20012563/cbb41382-a27d-11e6-92a9-18dab4fd1ad3.jpg -O person.jpg
+ ```
+
+Alternately, you can get the entire SSD Model artifacts + images in one single script from the MXNet Repository by running [get_ssd_data.sh script](https://github.com/apache/incubator-mxnet/blob/master/scala-package/examples/scripts/infer/objectdetector/get_ssd_data.sh)
+
+## Time to code!
+1. Following the [MXNet Java Setup on IntelliJ IDEA](/java/mxnet_java_on_intellij.html) tutorial, in the same project `JavaMXNet`, create a new empty class called : `ObjectDetectionTutorial.java`.
+2. In the `main` function of `ObjectDetectionTutorial.java` in let's define the downloaded model path and the image data paths in code. This is the same path where we downloaded the model artifacts and images as was shown above.
+
+ ```java
+ String modelPathPrefix = "/tmp/resnet50_ssd/resnet50_ssd_model";
+ String inputImagePath = "/tmp/resnet50_ssd/images/dog.jpg";
+ ```
+3. We can run the inference code in this example on either CPU or GPU (if you have a GPU backed machine) by choosing the appropriate context.
+
+ ```java
+
+ List<Context> context = getContext();
+ ...
+
+ private static List<Context> getContext() {
+ List<Context> ctx = new ArrayList<>();
+ ctx.add(Context.cpu()); // Choosing CPU Context here
+
+ return ctx;
+ }
+ ```
+4. To provide an input to the model, define the input shape to the model and the Input Data Descriptor (DataDesc) as shown below :
+
+ ```java
+ Shape inputShape = new Shape(new int[] {1, 3, 512, 512});
+ List<DataDesc> inputDescriptors = new ArrayList<DataDesc>();
+ inputDescriptors.add(new DataDesc("data", inputShape, DType.Float32(), "NCHW"));
+ ```
+
+ The input shape can be interpreted as follows : The input has a batch size of 1, with 3 RGB channels in the image, and the height and width of the image is 512 each.
+
+5. To run an actual inference on the given image, add the following lines to the `ObjectDetectionTutorial.java` class :
+
+ ```java
+ BufferedImage img = ObjectDetector.loadImageFromFile(inputImagePath);
+ ObjectDetector objDet = new ObjectDetector(modelPathPrefix, inputDescriptors, context, 0);
+ List<List<ObjectDetectorOutput>> output = objDet.imageObjectDetect(img, 3); // Top 3 objects detected will be returned
+ ```
+
+6. Let's piece all of the above steps together by showing the final contents of the `ObjectDetectionTutorial.java`.
+
+ ```java
+ package mxnet;
+
+ import org.apache.mxnet.infer.javaapi.ObjectDetector;
+ import org.apache.mxnet.infer.javaapi.ObjectDetectorOutput;
+ import org.apache.mxnet.javaapi.Context;
+ import org.apache.mxnet.javaapi.DType;
+ import org.apache.mxnet.javaapi.DataDesc;
+ import org.apache.mxnet.javaapi.Shape;
+
+ import java.awt.image.BufferedImage;
+ import java.util.ArrayList;
+ import java.util.Arrays;
+ import java.util.List;
+
+ public class ObjectDetectionTutorial {
+
+ public static void main(String[] args) {
+
+ String modelPathPrefix = "/tmp/resnet50_ssd/resnet50_ssd_model";
+
+ String inputImagePath = "/tmp/resnet50_ssd/images/dog.jpg";
+
+ List<Context> context = getContext();
+
+ Shape inputShape = new Shape(new int[] {1, 3, 512, 512});
+
+ List<DataDesc> inputDescriptors = new ArrayList<DataDesc>();
+ inputDescriptors.add(new DataDesc("data", inputShape, DType.Float32(), "NCHW"));
+
+ BufferedImage img = ObjectDetector.loadImageFromFile(inputImagePath);
+ ObjectDetector objDet = new ObjectDetector(modelPathPrefix, inputDescriptors, context, 0);
+ List<List<ObjectDetectorOutput>> output = objDet.imageObjectDetect(img, 3);
+
+ printOutput(output, inputShape);
+ }
+
+
+ private static List<Context> getContext() {
+ List<Context> ctx = new ArrayList<>();
+ ctx.add(Context.cpu());
+
+ return ctx;
+ }
+
+ private static void printOutput(List<List<ObjectDetectorOutput>> output, Shape inputShape) {
+
+ StringBuilder outputStr = new StringBuilder();
+
+ int width = inputShape.get(3);
+ int height = inputShape.get(2);
+
+ for (List<ObjectDetectorOutput> ele : output) {
+ for (ObjectDetectorOutput i : ele) {
+ outputStr.append("Class: " + i.getClassName() + "\n");
+ outputStr.append("Probabilties: " + i.getProbability() + "\n");
+
+ List<Float> coord = Arrays.asList(i.getXMin() * width,
+ i.getXMax() * height, i.getYMin() * width, i.getYMax() * height);
+ StringBuilder sb = new StringBuilder();
+ for (float c: coord) {
+ sb.append(", ").append(c);
+ }
+ outputStr.append("Coord:" + sb.substring(2)+ "\n");
+ }
+ }
+ System.out.println(outputStr);
+
+ }
+ }
+ ```
+
+7. To compile and run this code, change directories to this project's root folder then run the following:
Review comment:
```suggestion
7. To compile and run this code, change directories to this project's root folder, then run the following:
```
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
With regards,
Apache Git Services