update STARK-Lightning README

README.md

@@ -96,8 +96,6 @@ python tracking/train.py --script stark_st2 --config baseline --save_dir . --mod
 # STARK-ST101
 python tracking/train.py --script stark_st1 --config baseline_R101 --save_dir . --mode multiple --nproc_per_node 8  # STARK-ST101 Stage1
 python tracking/train.py --script stark_st2 --config baseline_R101 --save_dir . --mode multiple --nproc_per_node 8 --script_prv stark_st1 --config_prv baseline_R101  # STARK-ST101 Stage2
-# STARK-Lightning
-python tracking/train.py --script stark_lightning_X_trt --config baseline_rephead_4_lite_search5 --save_dir . --mode multiple --nproc_per_node 8  # STARK-Lightning
 ```
 (Optionally) Debugging training with a single GPU
 ```

install_pytorch17.sh

@@ -85,7 +85,7 @@ pip install git+https://github.com/votchallenge/vot-toolkit-python
 echo ""
 echo ""
 echo "****************** Installing onnx and onnxruntime-gpu ******************"
-pip install onnx onnxruntime-gpu==1.5.1
+pip install onnx onnxruntime-gpu==1.6.0
 
 echo ""
 echo ""

lib/tutorials/STARK_Lightning_Ch.md

@@ -1,6 +1,20 @@
 # STARK-Lightning 中文教程
 **前言**： [ONNXRUNTIME](https://github.com/microsoft/onnxruntime) 是微软开源的一个用于网络推理加速的库，在该教程中我们将教给大家如何将训练好的模型导出成ONNX格式，
-并使用ONNXRUNTIME来进一步加速推理，加速后的STARK-Lightning在RTX TITAN上的运行速度可达200~300 FPS！让我们开始吧
+并使用ONNXRUNTIME来进一步加速推理，加速后的STARK-Lightning在RTX TITAN上的运行速度可达200+ FPS！让我们开始吧
+## STARK-Lightning v.s 其他跟踪器
+| Tracker | LaSOT (AUC)| Speed (FPS) | Params (MB)|
+|---|---|---|---|
+|**STARK-Lightning**|**58.2**|**~200**|**8.2**|
+|DiMP50|56.8|~50|165|
+|DaSiamRPN|41.5|~200|362|
+|SiamFC|33.6|~100|8.9|
+STARK-Lightning取得了比DiMP50更强的性能，运行速度和DaSiamRPN一样快 :zap: ，而模型大小比SiamFC还要更小一些！
+## (非必须) 训练 STARK-Lightning
+运行下面的指令，可8卡并行训练
+```
+python tracking/train.py --script stark_lightning_X_trt --config baseline_rephead_4_lite_search5 --save_dir . --mode multiple --nproc_per_node 8
+```
+由于STARK-Lightning的训练很快，并且只需要极少的显存，因此也可以考虑用2卡或者4卡训练，只需对应修改 ```nproc_per_node```即可.
 ## 安装onnx和onnxruntime
 如果想在GPU上使用onnxruntime完成推理
 ```

lib/tutorials/STARK_Lightning_En.md

@@ -1,6 +1,20 @@
 # STARK-Lightning Tutorial
 **Introduction**：[ONNXRUNTIME](https://github.com/microsoft/onnxruntime) is an open-source library by Microsoft for network inference acceleration. In this tutorial, we will show how to export the trained model to ONNX format 
-and use ONNXRUNTIME to further accelerate the inference. The accelerated STARK-Lightning can run at 200~300 FPS on a RTX TITAN GPU! let's get started. 
+and use ONNXRUNTIME to further accelerate the inference. The accelerated STARK-Lightning can run at 200+ FPS on a RTX TITAN GPU! let's get started. 
+## STARK-Lightning v.s Other Trackers
+| Tracker | LaSOT (AUC)| Speed (FPS) | Params (MB)|
+|---|---|---|---|
+|**STARK-Lightning**|**58.2**|**~200**|**8.2**|
+|DiMP50|56.8|~50|165|
+|DaSiamRPN|41.5|~200|362|
+|SiamFC|33.6|~100|8.9|
+STARK-Lightning achieves better performance than DiMP50, runs at a competitive speed as DaSiamRPN, and has a smaller model size than SiamFC!
+## (Optionally) Train STARK-Lightning
+Train STARK-Lightning with 8 GPUs with the following command
+```
+python tracking/train.py --script stark_lightning_X_trt --config baseline_rephead_4_lite_search5 --save_dir . --mode multiple --nproc_per_node 8
+```
+Since the training of STARK-Lightning is fast and memory-friendly, you can also train it with less GPUs (such as 2 or 4) by set ```nproc_per_node``` accordingly.  
 ## Install onnx and onnxruntime
 for inference on GPU
 ```