wronk/divot-detect-inference repository overview

⁠Overview

This is a TensorFlow Serving image containing a model for detecting craters in Martian and Lunar satellite images (specifically, CTX and NAC images). The model uses the Mask RCNN architecture, which is an instance segmentation model. It generates crater masks as well as bounding box locations. For more information, see the github repository here⁠.

Sending images/Receiving predictions via REST The docker image takes in a json-like dict (via a POST request) with base64 encoded images as input. The input should be encoded as in the example below and all keywords much match exactly. You can include as many images as will fit in memory, but standard GPU machine are likely to hit out of memory errors after 2 images are included because the Mask RCNN model is quite heavy.

{
  "instances": [
    {
      "image_bytes": {"b64": "iVBO…Oxs6"}
    },
    {
      "image_bytes": {"b64": "0KGg…Pyg8"}
    },
    {
      "image_bytes": {"b64": "AABr…EKA0"}
    }
  ]
}

The model will return a json response with crater bounding boxes, confidences, and 33x33 pixel mask images.

⁠Running with docker

Run the GPU version like: docker run --runtime=nvidia -p 8501:8501 -t wronk/divot-detect-inference:v1-gpu

Make sure you have the requirements satisfied (GPU, NVIDIA drivers, and nvidia-docker) as outlined here⁠.