Haoping Bai

Dirichlet noise只在每次get action probability时的root node加，之后扩展的node是不加的。上面问题尝试弄清楚action selection时是在mcts前的prior probability加noise还是在mcts后用visit count算出的search probability加noise。前者的话是不会太影响搜索效果的，只会promote每次mcts playout时第一步尝试尽可能多的下子方法，不会影响之后的下子以及playout结束后使用最后的search probability来选择一个strong move，而后者则会直接在根据search probability选择action时引入noise，进而有可能undermine mcts作为“policy improvement process”的效果

Thanks for the quick response! The clarification helps a lot. One minor follow-up, the [`point_linear`](https://github.com/LaVieEnRoseSMZ/OQA/blob/99cd181280e97140e228f4d7fb7c71e8412c9aa1/elastic_nn/modules/dynamic_quant_layers.py#L57-L60) layer in MBConv has a linear activation, which would mean that the input to the...

I have a finetune starter code for COCO instance segmentation format data with some basic functionalities at [this repo](https://github.com/bhpfelix/segment-anything-finetuner). Hope it would help!

Thank you for the pointer! Congratulation on the strong results. The implementation here is currently not achieving desirable performance. The official implementation would be very helpful.

Hi, Sorry for the confusion, I haven't got a chance to maintain the project lately, and some of the documentation may be outdated. 699989 should be the number of the...