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

Why do we need to add tensor_label_k as input during SSOD？ #11

Open becauseofAI opened 1 year ago

becauseofAI commented 1 year ago

The code block is as follow: https://github.com/amazon-science/omni-detr/blob/main/engine.py#L184-L198

Why do we need to add tensor_label_k as input during SSOD？ When samples append tensor_label_q, tensor_label_k, tensor_unlabel_q as input, the cuda memory will increase until out of memery.

The CUDA out of memory information is as follows: RuntimeError: CUDA out of memory. Tried to allocate 506.00 MiB (GPU 1; 31.75 GiB total capacity; 27.74 GiB already allocated; 424.00 MiB free; 29.83 GiB reserved in total by PyTorch)

RuntimeError: CUDA out of memory. Tried to allocate 1.97 GiB (GPU 0; 79.35 GiB total capacity; 56.13 GiB already allocated; 1.38 GiB free; 57.79 GiB reserved in total by PyTorch)

peiwang062 commented 1 year ago

tensor_label_k is the weakly augmented image. We just followed the standard teacher-student framework. It is just one image, shouldn't be too big.