chore: yolox export filenames and documentation

tools/torch/README.md

@@ -16,3 +16,23 @@ Utility script to trace NLP models from Huggingface's pytorch-transformers. Requ
 pip3 install --user pytorch-transformers==1.1
 ```
 At the moment a CUDA model can not be convertible to CPU and vice versa. This may change with a future version of pytorch-transformers.
+
+* `trace_yolox.py`
+
+Requirements:
+
+```
+pip3 install thop
+pip3 install loguru
+```
+and custom YOLOX repository:
+```
+git clone https://github.com/jolibrain/YOLOX.git
+git checkout jit_script
+```
+
+Then to export a pretrained model before using it with DeepDetect:
+
+```
+python3 trace_yolox.py yolox-m -o /path/to/output/ --yolox_path /path/to/YOLOX/ --backbone_weights /path/to/pretrained/yolox_m.pth --num_classes 2 --img_width 300 --img_height 300
+```

tools/torch/trace_yolox.py

@@ -104,7 +104,7 @@ def main():
         model.to(device)
         model.eval()
 
-        filename += ".pt"
+        filename += "-" + str(args.img_width) + 'x' + str(args.img_height) + "_cls" + str(args.num_classes) + ".pt"
         script_module = torch.jit.script(model)
         logging.info("Save jit model at %s" % filename)
         script_module.save(filename)