where() missing 2 required positional argument: "input", "other"

[Chrispaoge]

您好，在跑train.py的时候遇到了这个错误 ： File "D:\code\pytorchCode\MCMOT\src\lib\trains\mot.py", line 198, in forward inds = torch.where(cls_id_map == cls_id)， TypeError: where() missing 2 required positional argument: "input", "other" 在计算ReID loss这一块，inds = torch.where(cls_id_map == cls_id)，我应该如何来改呢？来此where函数里，我应该填入哪两个参数呢？谢谢

[CaptainEven]

@Chrispaoge 首先，你要确保你的opts.py参数设置正确(cls_ids, cls2id, id2cls等)，然后关注训练的log, 比如hm的维数等。祝你好运！

[Chrispaoge]

@CaptainEven 您好，我觉得应该是我没表达清楚，您的代码在计算ReID损失时，用到了torch.where()函数，在这里您只填入了“ cls_id_map == cls_id“这一判断条件，即

torch.where()还缺少填入输入张量和输出张量，阅读代码后我还是不确定应该填入哪两个。不知道是不是我pytorch版本的问题，我用的pytorch1.1，已跑通您的FairMOTVehicle，但是我看了下pytorch1.1和高版本的pytorch中torch.where()函数并没有变化 cls_ids等参数都已设置好，但是torch.where()这个问题不解决，训练跑不起来

Traceback (most recent call last): File "D:/code/pytorchCode/MCMOT/src/train.py", line 162, in run(opt) File "D:/code/pytorchCode/MCMOT/src/train.py", line 121, in run log_dicttrain, = trainer.train(epoch, train_loader) File "D:\code\pytorchCode\MCMOT\src\lib\trains\base_trainer.py", line 162, in train return self.run_epoch('train', epoch, data_loader) File "D:\code\pytorchCode\MCMOT\src\lib\trains\base_trainer.py", line 97, in run_epoch output, loss, loss_stats = model_with_loss.forward(batch) File "D:\code\pytorchCode\MCMOT\src\lib\trains\base_trainer.py", line 24, in forward loss, loss_stats = self.loss.forward(outputs=outputs, batch=batch) File "D:\code\pytorchCode\MCMOT\src\lib\trains\mot.py", line 198, in forward inds = torch.where(cls_id_map == cls_id) TypeError: where() missing 2 required positional argument: "input", "other"

[CaptainEven]

@Chrispaoge 据我所知，torch.where函数跟numpy.where函数几乎一致，

并没有要求一定填入输入张量和输出张量，我用的是第二种方法，通过where获取索引，因此不需要填入x, y张量

[Ronales]

this problem may be caused by pytorch version, you can modify follow this：

inds = torch.where(cls_id_map == cls_id)

inds = (cls_id_map[:] == cls_id).nonzero().t()