search

maybeLx / MVSFormerPlusPlus

Codes of MVSFormer++: Revealing the Devil in Transformer’s Details for Multi-View Stereo (ICLR2024)
Apache License 2.0
153 stars 4 forks source link

error when using DDP mode #7

Open burui11087 opened 4 months ago

burui11087 commented 4 months ago

Hi, I'm having some issues with the training for blendedmvs using DDP mode.

`Traceback (most recent call last): File "train.py", line 265, in mp.spawn(main, nprocs=args.world_size, args=(args, config)) File "python3.8/site-packages/torch/multiprocessing/spawn.py", line 239, in spawn return start_processes(fn, args, nprocs, join, daemon, start_method='spawn') File "python3.8/site-packages/torch/multiprocessing/spawn.py", line 197, in start_processes while not context.join(): File "python3.8/site-packages/torch/multiprocessing/spawn.py", line 160, in join raise ProcessRaisedException(msg, error_index, failed_process.pid) torch.multiprocessing.spawn.ProcessRaisedException:

-- Process 2 terminated with the following error: Traceback (most recent call last): File "python3.8/site-packages/torch/multiprocessing/spawn.py", line 69, in _wrap fn(i, *args) File "MVSFormerPlusPlus/train.py", line 207, in main trainer.train() File "MVSFormerPlusPlus/base/base_trainer.py", line 79, in train result = self._train_epoch(epoch) File "MVSFormerPlusPlus/trainer/mvsformer_trainer.py", line 128, in _train_epoch outputs = self.model.forward(imgs_tmp, cam_params_tmp, depth_values[b_start:b_end]) File "python3.8/site-packages/torch/nn/parallel/distributed.py", line 1139, in forward if torch.is_grad_enabled() and self.reducer._rebuild_buckets(): RuntimeError: Expected to have finished reduction in the prior iteration before starting a new one. This error indicates that your module has parameters that were not used in producing loss. You can enable unused parameter detection by passing the keyword argument find_unused_parameters=True to torch.nn.parallel.DistributedDataParallel, and by making sure all forward function outputs participate in calculating loss. If you already have done the above, then the distributed data parallel module wasn't able to locate the output tensors in the return value of your module's forward function. Please include the loss function and the structure of the return value of forward of your module when reporting this issue (e.g. list, dict, iterable).`

Has anyone had the same issue? Thanks

Parskatt commented 3 months ago

This is probably caused by e.g. frozen parameters such as DINOv2 not getting gradients and DDP being stupid. What I have done in the past is just to put those in a list or something to prevent this type of error. You can also set find_unused_parameters=True like they suggest.