/instance-segmentation-deployment

Primary LanguagePython

Service deployment

Your task is to create an Instance Segmentation (more specifically, Instance Detection) service. You can use any pretrained model (such as these, for example). Instance segmentation metrics will not be counted in your final grade, but MAP should be above 0.5 in order to pass the tests.

Note that object names are the same as in the COCO 2017 dataset. You can retrieve them from the FasterRCNN_ResNet50_FPN_V2 model:

from torchvision.models.detection import FasterRCNN_ResNet50_FPN_V2_Weights

print(FasterRCNN_ResNet50_FPN_V2_Weights.COCO_V1.meta['categories'])

[4 points] HTTP endpoint: Implement a service which can handle the POST /predict query on port 8080. Request data is a JSON with the following structure:

{
  "url": "<url to image>"
}

Response data is also a JSON with the following structure:

{
  "objects": [
    "<object name>",
    "<object name>",
    ...
  ]
}

Example:

curl -XPOST http://localhost:8080/predict -H "Content-Type: application/json" -d '{"url": "http://images.cocodataset.org/val2017/000000001268.jpg"}' 
...
{
  "objects": [
        "bird",
        "boat",
        "boat",
        "person",
        "person",
        "person",
        "person",
        "cell phone",
        "backpack",
        "handbag",
        "boat"
    ]
}

[3 points] Metric endpoint: Implement a service which is able to serve its metrics in the Prometheus format via /metrics on the 8080 port. The most important metric for us is app_http_inference_count, the number of HTTP endpoint invocations.

Example:

curl http://localhost:8080/metrics
...
# HELP app_http_inference_count_total Multiprocess metric
# TYPE app_http_inference_count_total counter
app_http_inference_count_total 12.0

[3 points] GRPC endpoint: Implement a separate gRPC service on the 9090 port. See the inference.proto file in the proto directory. The contract is the same as for the HTTP endpoint.

How to submit?

  • Create a private GitHub repository
  • Install greater-solution-extractor-59 to your account and authorize it to your newly created repo.
  • Put Dockerfile in the root of the repository. This Dockerfile should assemble all code in your repo and the model checkpoint into the working service.
  • Go to http://week07.hareburrow.space:8080 , use student/Student!1 for login.
  • Build week07-pipeline with your repo as a parameter.
  • After successful build, click Keep this build forever
  • This link to your successful (or maybe only partially successful) is the solution for this homework - send it via anytask/LMS.
  • Also, add the link to your GitHub repo and the commit hash which corresponds to your submission.

Notes

  • Pretrained PyTorch model maskrcnn_resnet50_fpn with score_threshold = 0.75 works fine for this task
  • You are not limited to running only one process inside docker - use supervisord to run as many processes as you want
  • You can find actual tests in file tests.py
  • In order to sanity-check your model, tests check that MAP is greater than 0.5 - see tests for details