ChanganVR
commented
3 years ago The number of pedestrians can be set to a random number in the training as well as in the test. To obtain the best testing performance, I think the number of pedestrians in the training should be the same as the number of pedestrians in the test. But I haven't verified this.
I set the number of pedestrians to 5 because I needed to stick to one setup. But you can change that number to any numbers you want.
Hello I saw that you mentioned in your paper that you use the attention module to handle the variable number of pedestrians, but when looking at your code, I saw that you set the number of other pedestrians to 5, not see the treatment of different length. (See the padding code in explorer.py, but you commented them out) Therefore, do you assume a fixed number of pedestrians during training, so when you change the number of pedestrians, you need to retrain the model, or are there other things I haven't noticed? Looking forward to your answer 您好 不太清楚我的表达是不是很清楚，看到您似乎也是中国人，因此中文再说一遍：我看到在论文里您提到attention模块可以处理不同数量的行人，但是我在您的代码里是直接设置了行人数量为5。（我在explorer.py这个文件里看到了一些padding的代码，但是被注释掉了） 因此，请问您在训练中是假定了一个固定数量的行人么？那这样的话，如果拿到另外一个行人数量不通的环境里还需要重新train模型么？或者说还是有其他我没有注意到的地方呢？ 期待回复，祝科研顺利