search

leolyj / IRAST

This is the code for the ECCV2020 paper "Semi-Supervised Crowd Counting via Self-Training on Surrogate Tasks"
5 stars 3 forks source link

Labeled image should be multiplied by four #1

Open xuyushan opened 2 years ago

xuyushan commented 2 years ago

Hello, I'd like to ask you about the code. In the listDataset in dataset.py, if it is a training process, the labeled data should be multiplied by four. What's the reason for this?Looking forward to your reply.

Liuyan9527zns commented 2 years ago

Hi, Thanks for your attention to our work. We have two reasons to explain your question: 1) We follow the CSRNet to construct our IRAST framework. You can refer to the code: https://github.com/leeyeehoo/CSRNet-pytorch/blob/master/dataset.py, you can find this work also multiple the original images 4 times. 2) In this work, only a small number of images are labeled, for example, in ShanghaiTech Part A, the number of labeled images vs unlabeled images is 30 vs 210, we multiple the labeled images 4 times to mitigate the unbalanced between the training samples of labeled and unlabeled images.