search

amazon-science / omni-detr

PyTorch implementation of Omni-DETR for omni-supervised object detection: https://arxiv.org/abs/2203.16089
Other
64 stars 6 forks source link

code of forward propagation is incorrect #12

Open wwwbq opened 1 year ago

wwwbq commented 1 year ago

In the func of train_one_epoch_semi，the input of model is "sample", which is a list composed of tensors，the shape of tenosr is (b, c, h, w) , so tensor_list[0].ndim should be 4, but this repo only support the "if tensor_list[0].ndim == 3". I guess that the sample may be composed of some tensors which has the shape of (c, h, w), and len(sample) == batch ?

peiwang062 commented 1 year ago

The current code only support batch size=1. This if statement is from original deformableDETR implementation, we didn't make any change on it.