我只用一个GPU进行复现,相对应的调整了batch_size和学习率,复现结果如附图,感知,预测的指标都与论文中的相差不大,但是唯独规划指标相差很大。我的疑问是有没有人遇到和我一样的问题,如果有希望一起讨论解决办法。
(I only used one GPU for reproduction, and accordingly adjusted the batch size and learning rate. The reproduction results are shown in the attached figure. The indicators for perception and prediction are not much different from those in the paper, but the planning indicator differs greatly. My question is whether anyone has encountered the same problem as me. If so, I hope we can discuss and find a solution together.)