This model is designed accoding to the template in UniversalModelTemaplate. The model passes all the test cases in this application, and should work in the context of the server.
In the model directory, the code in model.py uses the pretrained FasterRCNN model in PyTorch to detect objects with prediction scores greater than 0.75.
This model is trained on the MS-COCO dataset which has 80 classes of objects (excluding background).
In model/coco_labels_super.json, these classes are grouped into 10 super-classes. The model returns the number of objects detected in each of the super-classes.
| Super-class | COCO Class Label |
|---|---|
| person | person |
| modes of transport | bicycle, car, motorcycle, airplane, bus, train, truck, boat |
| street view | traffic light, fire hydrant, stop sign, parking meter, bench |
| animals | bird, cat, dog, horse, sheep, cow, elephant, bear, zebra, giraffe |
| sports | frisbee, skis, snowboard, sports ball, kite, baseball bat, baseball glove, skateboard, surfboard, tennis racket |
| food | banana, apple, sandwich, orange, broccoli, carrot, hot dog, pizza, donut, cake |
| kitchen | wine glass, cup, fork, knife, spoon, bowl |
| indoor | couch, potted plant, bed, dining table, toilet, sink, clock, vase |
| electronis | tv, laptop, mouse, remote, keyboard, cell phone, microwave, oven, toaster, refrigerator, hair drier |
| misc | book, scissors, teddy bear, toothbrush, tie, backpack, umbrella, handbag, suitcase, chair, bottle |
It is sufficient to make any required modifications or regrouping of the class labels in the JSON file itself. Everywhere else the changes would follow.
config.py has some metadata about the ML model, e.g. input type, model name, and tags.
The requirements are added to requirements.py file in the model directory.
In the root directory of the project, run the command
docker-compose build debug and then docker-compose up debug. Open a web browser and navigate to
http://localhost:4650.
As you make changes to your model the results will appear on the web page showing the initialization status and a prediction result on a test image.
In the root directory of the project, run the command docker-compose build test and then
docker-compose up test. You will see the results of the test cases in your terminal. If all
test cases pass, then your model will work in the server environment.