tacju / compositor Goto Github PK
View Code? Open in Web Editor NEWThis repo contains the code for our paper Compositor: Bottom-Up Clustering and Compositing for Robust Part and Object Segmentation
License: Apache License 2.0
This repo contains the code for our paper Compositor: Bottom-Up Clustering and Compositing for Robust Part and Object Segmentation
License: Apache License 2.0
The pascal part configuration and dataset could not be found and the transformer based configuration could not be used because there was no bs16_90k configuration and the KMax-DeepLab folder was not found
Followed the install and dataset instruction set without issues. However, when trying to run Compositor and its variants train_net.py
crashes before or during training iteration 0. Examples provided below:
python train_net.py --num-gpus 4 --config-file configs/partimagenet/semantic-segmentation/swin/compositor_swin_tiny_bs16_90k.yaml
-- Process 1 terminated with the following error:
FileNotFoundError: [Errno 2] No such file or directory: 'configs/partimagenet/semantic-segmentation/compositor_R50_bs16_90k.yaml'
python train_net.py --num-gpus 4 --config-file configs/partimagenet/semantic-segmentation/compositor_R50_bs128_40k.yaml
-- Process 3 terminated with the following error:
File "/projects/jare1686/software/anaconda/envs/compositor/lib/python3.9/site-packages/torch/_tensor.py", line 972, in __array__
return self.numpy().astype(dtype, copy=False)
TypeError
python train_net.py --num-gpus 4 --config-file configs/partimagenet/semantic-segmentation/Base-PartImageNet-SemanticSegmentation.yaml
-- Process 2 terminated with the following error:
raise ValueError("Cannot match one checkpoint key to multiple keys in the model.")
ValueError: Cannot match one checkpoint key to multiple keys in the model.
Any help is greatly appreciated.
Hello,
I have certain hardware constraints. And need to run this code on cuda11.8. However, getting issues with MultiScaleDeformableAttention.
Would it be possible to resolve this issue? Please let me know if you are aware to solve this problem.
Error: import MultiScaleDeformableAttention as MSDA ImportError: libtorch_cuda_cu.so: cannot open shared object file: No such file or directory
Thanks
Hello!
With the code and configs provided I am unable to reproduce the Compositor MaskFormer results on PartImageNet. I've tried a few different learning rates but the best results I've observed are ~50 mIoU for part and ~62 for object. Could you release checkpoints and training logs?
Thank you!
Josh
Hi,
Would it be possible to release the Pascal Part datasets as well?
Thanks
Hello!
Thank you for releasing this and with thorough instructions! I just went to try training the DeepLab variant (with 1 gpu) and got the following error
Traceback (most recent call last):
File "/projects/jomy5901/Compositor/Compositor_kMaX-DeepLab/train_net.py", line 334, in <module>
launch(
File "/projects/jomy5901/Compositor/Compositor_kMaX-DeepLab/tmp/detectron2/detectron2/engine/launch.py", line 84, in launch
main_func(*args)
File "/projects/jomy5901/Compositor/Compositor_kMaX-DeepLab/train_net.py", line 328, in main
return trainer.train()
File "/projects/jomy5901/Compositor/Compositor_kMaX-DeepLab/tmp/detectron2/detectron2/engine/defaults.py", line 486, in train
super().train(self.start_iter, self.max_iter)
File "/projects/jomy5901/Compositor/Compositor_kMaX-DeepLab/tmp/detectron2/detectron2/engine/train_loop.py", line 155, in train
self.run_step()
File "/projects/jomy5901/Compositor/Compositor_kMaX-DeepLab/tmp/detectron2/detectron2/engine/defaults.py", line 496, in run_step
self._trainer.run_step()
File "/projects/jomy5901/Compositor/Compositor_kMaX-DeepLab/tmp/detectron2/detectron2/engine/train_loop.py", line 494, in run_step
loss_dict = self.model(data)
File "/projects/jomy5901/software/anaconda/envs/compositor/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, **kwargs)
File "/projects/jomy5901/Compositor/Compositor_kMaX-DeepLab/compositor/compositor_model.py", line 232, in forward
losses = self.criterion(outputs, targets)
File "/projects/jomy5901/software/anaconda/envs/compositor/lib/python3.9/site-packages/torch/nn/modules/module.py", line 1051, in _call_impl
return forward_call(*input, **kwargs)
File "/projects/jomy5901/Compositor/Compositor_kMaX-DeepLab/compositor/modeling/criterion.py", line 386, in forward
losses.update(self.get_loss(loss, outputs, processed_targets, prefix='part_'))
File "/projects/jomy5901/Compositor/Compositor_kMaX-DeepLab/compositor/modeling/criterion.py", line 283, in get_loss
return loss_map[loss](outputs, targets, prefix=prefix)
File "/projects/jomy5901/Compositor/Compositor_kMaX-DeepLab/compositor/modeling/criterion.py", line 217, in loss_labels
losses = {prefix + "loss_ce": focal_cross_entropy_loss(src_logits, target_classes, pq_loss_class_weight)}
File "/projects/jomy5901/Compositor/Compositor_kMaX-DeepLab/compositor/modeling/criterion.py", line 46, in focal_cross_entropy_loss
loss = F.cross_entropy(pred, gt, reduction="none") # B x N
File "/projects/jomy5901/software/anaconda/envs/compositor/lib/python3.9/site-packages/torch/nn/functional.py", line 2824, in cross_entropy
return torch._C._nn.cross_entropy_loss(input, target, weight, _Reduction.get_enum(reduction), ignore_index)
RuntimeError: Expected target size [16, 30], got [16, 42, 30]
Do you know why this may occur?
Thank you!
Hi! I'm having some problems when trying to install the dependencies for Mask2Former. I followed the installation instructions, but still got some errors. Do you have an alternative way of installing this library?
Thanks for the help
Hi,
Would it be possible to release the Pascal Part configs and datasets as well?
Thanks
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.