[Refactor] Refactor MMTracking_Tutorial.ipynb (#712)

* update tutorial and set randomness for sot * update tutorial * update lib install * update tutorial * fix mot dataset convert * refactor MMTracking_Tutorials * fix some version bugs * update the tutorial * update stark docs about the adamw optimizer
open-mmlab · Sep 7, 2022 · 814376d · 814376d
1 parent 94516ba
commit 814376d
Show file tree

Hide file tree

Showing 3 changed files with 2,732 additions and 2,349 deletions.
diff --git a/configs/sot/stark/README.md b/configs/sot/stark/README.md
@@ -80,6 +80,8 @@ The STARK is trained in 2 stages. We denote the 1st-stage model as `STARK-ST1`,
 
 Models from the 2 stages have different configurations. For example, `stark-st1_r50_8xb16-500e_got10k.py` is the configuration of the 1st-stage model and `stark-st2_r50_8xb16-50e_got10.py` is the configuration of the 2nd-stage model. The following is an example of training STARK on GOT10k dataset. The training on LaSOT and TrackingNet is similar like this.
 
+**Note** that the STARK can not be trained in the 1.8.0 version of pytorch since the potential [bug](https://github.com/pytorch/pytorch/pull/52944/files) about AdamW optimizer in pytorch.
+
 **Training the 1st-stage model**
 
 ```shell