feat: add yolox export script for training and inference

[Bycob]

Requires a modified version of the model available here: https://github.com/jolibrain/YOLOX/tree/jit_script

TODO:

• inference example
• training example
• Move modified repo to github/jolibrain
• Change template from "fasterrcnn" to something else

Inference example

git clone --branch jit_script https://github.com/jolibrain/YOLOX [YOLOX_LOCATION]
pip3 install loguru thop opencv-python tabulate

# from dede root
# download weights from YOLOX readme https://github.com/jolibrain/YOLOX
python3 tools/torch/trace_yolox.py -v yolox-s --yolox_path [YOLOX_LOCATION] -o /opt/platform/models/training/yolox_coco --weights yolox_s.pth --num_classe 80

# create service
curl -X PUT 'http://localhost:8500/services/yolox_coco' -d '{
    "description": "JIT YoloX on coco dataset",
    "mllib": "torch",
    "model": {
        "repository": "/opt/platform/models/training/yolox_coco/"
    },
    "parameters": {
        "input": {
            "connector": "image",
            "width":640,
            "height":640,
            "rgb":true
        },
        "mllib":{
            "template":"yolox",
            "gpu":true,
            "gpuid":0,
            "num_classes":80
        },
        "output": {
            "bbox": true
        }
    },
    "mltype": "detection",
    "type": "supervised"
}'

echo \

curl -X POST 'http://localhost:8500/predict/' -d '{
 "service": "yolox_coco",
 "parameters": {
  "input": {},
  "output": {
   "bbox":true,
   "confidence_threshold": 0.5
  },
  "mllib": {
   "gpu": true
  }
 },
 "data": [
  "/opt/platform/data/images/cat.jpg"
 ]
}'

Training example

# export model with a pretrained backbone
python3 tools/torch/trace_yolox.py -v yolox-s --yolox_path [YOLOX_LOCATION] -o /opt/platform/models/training/yolox_cars --backbone_weights yolox_s.pth --num_classe 2

# create service
curl -X PUT 'http://localhost:8500/services/yolox_cars' -d '{                                                                                                                                      
    "description": "Torch traced YoloX on cars with pretrained backbone",
    "mllib": "torch",
    "model": {
        "repository": "/opt/platform/models/training/yolox_cars/"
    },
    "parameters": {
        "input": {
            "connector": "image",
            "width":512,
            "height":512,
            "rgb":true,
            "bbox":true,
            "db":true
        },
        "mllib":{
            "template":"yolox",
            "nclasses":2,
            "gpu":true,
            "gpuid":0
        },
        "output": {
            "bbox": true,
            "store_config": true
        }
    },
    "mltype": "detection",
    "type": "supervised"
}'

# run training
curl -X POST 'http://localhost:8500/train' -d '{
   "service":"yolox_cars",
   "parameters":{
      "input":{
         "db":true
      },
      "output":{
         "measure":[
            "map"
         ]
      },
      "mllib":{
         "solver":{
            "iterations":15000,
            "base_lr":1e-4,
            "finetune":true,
            "iter_size":4,
            "solver_type":"ADAM",
            "test_interval":100,
            "snapshot": 2000
         },
         "net":{
            "batch_size":8
         },
         "mirror":true,
         "rotate":true,
         "geometry":0.5,
         "zoom_in":true,
         "persp_horizontal":true,
         "persp_vertical":true
      }
   },
   "data":[
      "/opt/platform/examples/cars/train.txt",
      "/opt/platform/examples/cars/test.txt"
   ]
}'