xingyuuchen / tri-depth Goto Github PK

Hello, thank you very much for your contribution. I get the following error when adding triplet loss to another network: Traceback (most recent call last):
File "train.py", line 17, in
trainer.train()
File "/root/autodl-tmp/DIFFNet/trainer.py", line 179, in train
self.run_epoch()
File "/root/autodl-tmp/DIFFNet/trainer.py", line 195, in run_epoch
outputs, losses = self.process_batch(inputs)
File "/root/autodl-tmp/DIFFNet/trainer.py", line 253, in process_batch
losses = self.compute_losses(inputs, outputs)
File "/root/autodl-tmp/DIFFNet/trainer.py", line 508, in compute_losses
sgt_loss = self.compute_sgt_loss(inputs, outputs)
File "/root/autodl-tmp/DIFFNet/trainer.py", line 516, in compute_sgt_loss
seg_target = inputs[('seg', 0, 0)]
KeyError: ('seg', 0, 0)
There is no ('seg',0,0) key in inputs.How to solve such a problem? Looking forward to your response.

Your outstanding work is amazing. I tried to use your proposed loss for monodepth2, but there are always bugs when importing data, can you provide the code for monodepth2 about kitti_dataset and mono_datasets, thank you.

Original Traceback (most recent call last):
File "/home/yzhang/anaconda3/envs/mono/lib/python3.8/site-packages/torch/utils/data/_utils/worker.py", line 202, in _worker_loop
data = fetcher.fetch(index)
File "/home/yzhang/anaconda3/envs/mono/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 44, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/home/yzhang/anaconda3/envs/mono/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 44, in
data = [self.dataset[idx] for idx in possibly_batched_index]
File "/home/yzhang/monodepth2-master/datasets/kitti_dataset.py", line 176, in getitem
self.preprocess(inputs, color_aug)
File "/home/yzhang/monodepth2-master/datasets/kitti_dataset.py", line 121, in preprocess
self.a=color_aug(f)
TypeError: 'tuple' object is not callable

RT，thanks

we find some error when we use your code to training our model，

‘seg_target = inputs[('seg', 0, 0)]‘show that there is no ’seg‘ in the inputs.

we notice that your code is based on FSRE and manydepth, so we copy the FSRE dataloader to tri-depth and debug the code, After making modifications to the data loader, we were able to successfully train our model using the provided code. We are curious to know if there are alternative methods for utilizing this code that we could explore

Thanks for releasing your amazing work. I am wondering if the pre-trained model is for ManyDepth?

Hi, author. Thanks for your remarkable work.

Dear author, the segmented picture provided in FSRE-Depth is gray, may I ask how the color segmented picture you provided in the article was obtained? I want to know how to get it, thank you!

Hello, it seems that the pre-trained model link cannot be downloaded

Hi, thanks for your nice work!

I'm wondering have you tested the results of Tri-depth with high-resolution (320*1024) inputs? It would be helpful if you could provide the related information or results.

Looking forward to your reply!

Hi @xingyuuchen, thank you for open-sourcing your work!

I would like to ask if it is possible for you to release the pretrained weights of TriDepth on KITTI for the purpose of reproduction. Thanks!