the-learning-and-vision-atelier-lava / posfeat Goto Github PK
View Code? Open in Web Editor NEW[CVPR2022] Decoupling Makes Weakly Supervised Local Feature Better
License: Apache License 2.0
[CVPR2022] Decoupling Makes Weakly Supervised Local Feature Better
License: Apache License 2.0
Hello, good job for PoseFeat. My work requires descriptor networks. Can you please provide the pre-trained model, thanks!
你好,我想问下我在评估ETH数据集的时候,extract_ETH.yaml中的colmap_path是自己建立的新文件夹吗?我建立新文件夹后,报错是:
Traceback (most recent call last):
File "reconstruction_pipeline.py", line 364, in
main()
File "reconstruction_pipeline.py", line 314, in main
matching_stats = import_features_and_match(configs, paths)
File "reconstruction_pipeline.py", line 146, in import_features_and_match
"--match_type", "pairs"])
File "/home/wujunke/anaconda3/envs/yolov7/lib/python3.7/subprocess.py", line 339, in call
with Popen(*popenargs, **kwargs) as p:
File "/home/wujunke/anaconda3/envs/yolov7/lib/python3.7/subprocess.py", line 800, in init
restore_signals, start_new_session)
File "/home/wujunke/anaconda3/envs/yolov7/lib/python3.7/subprocess.py", line 1551, in _execute_child
raise child_exception_type(errno_num, err_msg, err_filename)
PermissionError: [Errno 13] Permission denied: Path('/home/wujunke/Installed/colmap/build/src/exe/colmap')
请问为什么PosFeat的detector网络输入的是图像+fine特征图。而且在计算过程中计算相邻点的SSIM,最后在与对应的图像/特征图相乘,这样的目的是什么?可否理解为一种positional encoding,引入位置信息?谢谢!
Hi! I've been trying to reproduce your work, but I found trouble judging the convergence of the network, especially the DetNet.
I've learned from your previous reply that we should judge convergence with the score map when training the detector. My question is how to judge convergence from score maps. If the lines in score maps are sharp and clear, does it mean the detector has converged? Besides, how about the convergence of Descriptor? Can we use loss or should we use score maps?
Thanks a lot!
Hello, your work is very good, recently I have been reproducing your work, but the Viewpoint on the Hpatch dataset has never been able to achieve the effect in your paper, did you encounter such a problem at that time?
Hey,
I have trained the model, now I wanted to extract key points and descriptors.
For that you have provided us with extract.py
file but not with './configs/extract.yaml'
. Can you please provide default config for the same!
Thanks!
@KunB-Fighting
I fixed the issue in #4 by using kp1_comb = (int(kp1[0]+w+mid_pad), int(kp1[1]))
instead of kp1 in trainer.py
After doing that, the above error was fixed but it gave me a new following error:
line 1225, in _process_data
data.reraise()
File "/mnt/private/atharv/anaconda3/envs/posfeat/lib/python3.6/site-packages/torch/_utils.py", line 429, in reraise
raise self.exc_type(msg)
IndexError: Caught IndexError in DataLoader worker process 3.
Original Traceback (most recent call last):
File "/mnt/private/atharv/anaconda3/envs/posfeat/lib/python3.6/site-packages/torch/utils/data/_utils/worker.py", line 202, in _worker_loop
data = fetcher.fetch(index)
File "/mnt/private/atharv/anaconda3/envs/posfeat/lib/python3.6/site-packages/torch/utils/data/_utils/fetch.py", line 47, in fetch
return self.collate_fn(data)
File "/mnt/private/atharv/hyperturing/posfeat/PoSFeat/managers/trainer.py", line 151, in my_collate
return torch.utils.data.dataloader.default_collate(batch)
File "/mnt/private/atharv/anaconda3/envs/posfeat/lib/python3.6/site-packages/torch/utils/data/_utils/collate.py", line 45, in default_collate
elem = batch[0]
IndexError: list index out of range
not sure why batch[0] is out of index
Hi,
your work is very interesting. I was wondering if you could release the pretrained weights?
Thank you for releasing your code.
It seems that the reward loss in your code (line 179) is different from Equation (11) of the paper.
In your code (line 179), L_rew = P_m * R * log(P_m*P_kp(x)*P_kp(y)).
But the reward loss shown in Equation (11) of the paper is L_rew = P_m * R * log(P_kp(x)*P_kp(y)).
Maybe I am wrong.
Looking forward to hearing from you.
Hello, is there a pre-trained model available?
]
total number of image pairs loaded: 788921
descriptor/descriptor 1/10: 0%| | 0/10000 [00:03<?, ?it/s]
Traceback (most recent call last):
File "/home/swordwu/workspace/PoSFeat/train.py", line 9, in
trainer.train()
File "/home/swordwu/workspace/PoSFeat/managers/trainer.py", line 290, in train
self.val_and_vis(batch_path_list, idx)
File "/home/swordwu/miniconda3/envs/nerf_py39/lib/python3.9/site-packages/torch/autograd/grad_mode.py", line 27, in decorate_context
return func(*args, **kwargs)
File "/home/swordwu/workspace/PoSFeat/managers/trainer.py", line 499, in val_and_vis
cv2.circle(comb_img_kps, kp1, radius=2, color=color, thickness=-1)
cv2.error: OpenCV(4.5.5) 👎 error: (-5:Bad argument) in function 'circle'
Overload resolution failed:
- Can't parse 'center'. Sequence item with index 0 has a wrong type
- Can't parse 'center'. Sequence item with index 0 has a wrong type
Hi! I'm looking for a simple demo on how to extract features / keypoints from a single image with your model.
Could you provide this, or give a hint on where to get started? I would prefer not to write config files etc.
Hey,
Great work with PoSFeat, I want to replace SuperPoint in SuperGlue with your model, is the possible and which part should I use of your model to replace it.
Also can you please provide the pretrained model.
Thanks! :)
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 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. 📊📈🎉
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 ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.