Closed zhuhu00 closed 1 year ago
Hi, you can make the value of max_images
smaller. Sorry I should make it clearer. max_images
actually takes the role of batch size in our code, because we use a dynamic batch size. In the code below, you can see n_seqs = self.max_images // n_per_seq
. https://github.com/facebookresearch/PoseDiffusion/blob/8ce281338d46561f8a82b6f3a7675ad1712fc12f/pose_diffusion/util/train_util.py#L27-L59
Hi, Jianyuan, I tried to retrain in co3dv2, but there was an “out of memory” error in only one or two categories. Is there any way to solve it? my device is RTX 3090 with 24GB memory.