yw0208 / STAF

STAF: 3D Human Mesh Recovery from Video with Spatio-Temporal Alignment Fusion
MIT License
47 stars 4 forks source link

I met another question during the training #5

Open lisaner000 opened 6 months ago

lisaner000 commented 6 months ago

Traceback (most recent call last): File "train.py", line 168, in main(cfg) File "train.py", line 159, in main debug_freq=cfg.DEBUG_FREQ, File "/root/autodl-tmp/STAF-main/lib/core/trainer.py", line 400, in fit self.train() # 调用 train 方法进行模型训练 File "/root/autodl-tmp/STAF-main/lib/core/trainer.py", line 227, in train scores=scores, File "/root/miniconda3/envs/yzy_staf/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1110, in _call_impl return forward_call(*input, **kwargs) File "/root/autodl-tmp/STAF-main/lib/core/loss.py", line 116, in forward pred_theta = pred_theta[w_smpl] IndexError: The shape of the mask [39] at index 0 does not match the shape of the indexed tensor [0, 85] at index 0 Can you give me some advice?

yw0208 commented 6 months ago

It seems like there are some mistakes in data loading. Because it's wired, the shape of pred_theta is [0, 85]. You should debug to check what went wrong.

lisaner000 commented 6 months ago

It seems like there are some mistakes in data loading. Because it's wired, the shape of pred_theta is [0, 85]. You should debug to check what went wrong.

It achieves the training successfully. Thank you very much!

lisaner000 commented 6 months ago

Another, I met a bug during the evaluting as the following: Traceback (most recent call last): File "evaluate.py", line 94, in model.load_state_dict(checkpoint['gen_state_dict']) File "/root/miniconda3/envs/yzy_staf/lib/python3.7/site-packages/torch/nn/modules/module.py", line 1498, in load_state_dict self.class.name, "\n\t".join(error_msgs))) RuntimeError: Error(s) in loading state_dict for SAFM: Missing key(s) in state_dict: "nonlocalblock.attention.fc.weight", "nonlocalblock.attention.fc.bias", "nonlocalblock.attention.attention.0.weight", "nonlocalblock.attention.attention.0.bias", "nonlocalblock.attention.attention.2.weight", "nonlocalblock.attention.attention.2.bias", "nonlocalblock.attention.attention.4.weight", "nonlocalblock.attention.attention.4.bias". Unexpected key(s) in state_dict: "points_grid", "deconv_layers.0.weight", "deconv_layers.1.weight", "deconv_layers.1.bias", "deconv_layers.1.running_mean", "deconv_layers.1.running_var", "deconv_layers.1.num_batches_tracked", "deconv_layers.3.weight", "deconv_layers.4.weight", "deconv_layers.4.bias", "deconv_layers.4.running_mean", "deconv_layers.4.running_var", "deconv_layers.4.num_batches_tracked", "deconv_layers.6.weight", "deconv_layers.7.weight", "deconv_layers.7.bias", "deconv_layers.7.running_mean", "deconv_layers.7.running_var", "deconv_layers.7.num_batches_tracked", "maf_extractor.0.Dmap", "maf_extractor.0.conv0.weight", "maf_extractor.0.conv0.bias", "maf_extractor.0.conv1.weight", "maf_extractor.0.conv1.bias", "maf_extractor.0.conv2.weight", "maf_extractor.0.conv2.bias", "maf_extractor.1.Dmap", "maf_extractor.1.conv0.weight", "maf_extractor.1.conv0.bias", "maf_extractor.1.conv1.weight", "maf_extractor.1.conv1.bias", "maf_extractor.1.conv2.weight", "maf_extractor.1.conv2.bias", "maf_extractor.2.Dmap", "maf_extractor.2.conv0.weight", "maf_extractor.2.conv0.bias", "maf_extractor.2.conv1.weight", "maf_extractor.2.conv1.bias", "maf_extractor.2.conv2.weight", "maf_extractor.2.conv2.bias", "regressor.0.init_pose", "regressor.0.init_shape", "regressor.0.init_cam", "regressor.0.fc1.weight", "regressor.0.fc1.bias", "regressor.0.fc2.weight", "regressor.0.fc2.bias", "regressor.0.decpose.weight", "regressor.0.decpose.bias", "regressor.0.decshape.weight", "regressor.0.decshape.bias", "regressor.0.deccam.weight", "regressor.0.deccam.bias", "regressor.0.smpl.betas", "regressor.0.smpl.global_orient", "regressor.0.smpl.body_pose", "regressor.0.smpl.faces_tensor", "regressor.0.smpl.v_template", "regressor.0.smpl.shapedirs", "regressor.0.smpl.J_regressor", "regressor.0.smpl.posedirs", "regressor.0.smpl.parents", "regressor.0.smpl.lbs_weights", "regressor.0.smpl.J_regressor_extra", "regressor.0.smpl.vertex_joint_selector.extra_joints_idxs", "regressor.1.init_pose", "regressor.1.init_shape", "regressor.1.init_cam", "regressor.1.fc1.weight", "regressor.1.fc1.bias", "regressor.1.fc2.weight", "regressor.1.fc2.bias", "regressor.1.decpose.weight", "regressor.1.decpose.bias", "regressor.1.decshape.weight", "regressor.1.decshape.bias", "regressor.1.deccam.weight", "regressor.1.deccam.bias", "regressor.1.smpl.betas", "regressor.1.smpl.global_orient", "regressor.1.smpl.body_pose", "regressor.1.smpl.faces_tensor", "regressor.1.smpl.v_template", "regressor.1.smpl.shapedirs", "regressor.1.smpl.J_regressor", "regressor.1.smpl.posedirs", "regressor.1.smpl.parents", "regressor.1.smpl.lbs_weights", "regressor.1.smpl.J_regressor_extra", "regressor.1.smpl.vertex_joint_selector.extra_joints_idxs", "regressor.2.init_pose", "regressor.2.init_shape", "regressor.2.init_cam", "regressor.2.fc1.weight", "regressor.2.fc1.bias", "regressor.2.fc2.weight", "regressor.2.fc2.bias", "regressor.2.decpose.weight", "regressor.2.decpose.bias", "regressor.2.decshape.weight", "regressor.2.decshape.bias", "regressor.2.deccam.weight", "regressor.2.deccam.bias", "regressor.2.smpl.betas", "regressor.2.smpl.global_orient", "regressor.2.smpl.body_pose", "regressor.2.smpl.faces_tensor", "regressor.2.smpl.v_template", "regressor.2.smpl.shapedirs", "regressor.2.smpl.J_regressor", "regressor.2.smpl.posedirs", "regressor.2.smpl.parents", "regressor.2.smpl.lbs_weights", "regressor.2.smpl.J_regressor_extra", "regressor.2.smpl.vertex_joint_selector.extra_joints_idxs", "safm.nonlocalblock.attention.fc.weight", "safm.nonlocalblock.attention.fc.bias", "safm.nonlocalblock.attention.attention.0.weight", "safm.nonlocalblock.attention.attention.0.bias", "safm.nonlocalblock.attention.attention.2.weight", "safm.nonlocalblock.attention.attention.2.bias", "safm.nonlocalblock.attention.attention.4.weight", "safm.nonlocalblock.attention.attention.4.bias", "tcfm.nonlocalblock.conv_phi.weight", "tcfm.nonlocalblock.conv_theta.weight", "tcfm.nonlocalblock.conv_g.weight", "tcfm.nonlocalblock.conv_mask.weight", "tcfm.nonlocalblock.conv_mask_forR.weight".

yw0208 commented 6 months ago

refer to issue