chore(demo): update megengine deployment (#153)

README.md

@@ -8,6 +8,7 @@ For more details, please refer to our [report on Arxiv](https://arxiv.org/abs/21
 <img src="assets/git_fig.png" width="1000" >
 
 ## Updates!!
+* 【2021/07/26】 We now support [MegEngine](https://github.com/Megvii-BaseDetection/YOLOX/tree/main/demo/MegEngine) deployment.
 * 【2021/07/20】 We have released our technical report on [Arxiv](https://arxiv.org/abs/2107.08430).
 
 ## Comming soon

demo/MegEngine/cpp/yolox.cpp

@@ -15,8 +15,8 @@
 /**
  * @brief Define names based depends on Unicode path support
  */
-#define NMS_THRESH 0.65
-#define BBOX_CONF_THRESH 0.3
+#define NMS_THRESH 0.45
+#define BBOX_CONF_THRESH 0.25
 
 constexpr int INPUT_W = 640;
 constexpr int INPUT_H = 640;

demo/MegEngine/python/README.md

@@ -21,7 +21,7 @@ python3 convert_weights.py -w yolox_s.pth.tar -o yolox_s_mge.pkl
 This part is the same as torch's python demo, but no need to specify device.
 
 ```
-python3 demo.py image -n yolox-s -c yolox_s_mge.pkl --path ../../../assets/dog.jpg --conf 0.3 --nms 0.65 --tsize 640 --save_result
+python3 demo.py image -n yolox-s -c yolox_s_mge.pkl --path ../../../assets/dog.jpg --conf 0.25 --nms 0.45 --tsize 640 --save_result
 ```
 
 ###  [Optional]Step4: dump model for cpp inference