maunzzz / fine-grained-segmentation-networks Goto Github PK
View Code? Open in Web Editor NEWCode for the ICCV paper "Fine-Grained Segmentation Networks: Self-Supervised Segmentation for Improved Long-Term Visual Localization"
License: Other
Code for the ICCV paper "Fine-Grained Segmentation Networks: Self-Supervised Segmentation for Improved Long-Term Visual Localization"
License: Other
in your model folder for pspnet.py in line 194
if self.training:
if self.output_all:
aux = self.conv6_1relu(aux_feat)
aux = self.conv6_1(aux)
h = self.conv6relu(h_feat)
h = self.conv6(h)
return h_feat, aux_feat, h, aux
elif self.output_features:
return h_feat, aux_feat
else:
aux_feat = self.conv6_1relu(aux_feat)
aux = self.conv6_1(aux_feat)
h_feat = self.conv6relu(h_feat)
h = self.conv6(h_feat)
return aux, h------------this one, maybe return h, aux
when you are training ,you want to calculate the resnet loss and auxiliary loss from resnet and auxiliary layer. But in here you return in a wrong oder. At the end auxiliary loss will be the main loss for you loss function instead of resnet loss.
outputs_ref, aux_ref = net(img_ref) ---- line409 in train_with_clustering.py
.....
main_loss, _ = loss_fct(outputs_ref, outputs_other, None, pts_ref,
pts_other, cluster_labels=cluster_labels) ---- line437
Hi,
you have mentioned in you paper, that you integrate your segmentation result into this semantic visual localization algorithum (http://openaccess.thecvf.com/content_ECCV_2018/papers/Carl_Toft_Semantic_Match_Consistency_ECCV_2018_paper.pdf ) and use it to evaluation your performance as well. But I couldn't find the release code based on this localization method. Do you reproduct this paper on your own? And Is it possible for you to release the code for evaluation as well? Thank you in advance.
Can you give me the link of the correspondence datasets
Thanks for your great work!
I get errors after run this instruction:
run clustering/setup_cluster_dataset.py
the errors as follows:
Traceback (most recent call last):
File "/home/zxm/Test/fine-grained-segmentation-networks/clustering/setup_cluster_dataset.py", line 7, in
write_lists_for_corr(corr_set)
File "/home/zxm/Test/fine-grained-segmentation-networks/clustering/utils/write_lists_for_corr.py", line 20, in write_lists_for_corr
f_name_list = [fn for fn in os.listdir(corr_set_config.correspondence_path) if fn.endswith('mat')]
FileNotFoundError: [Errno 2] No such file or directory: 'dummy/path'
Thank you!
Hi,
Thanks for your great work!
I get errors after run the instruction:
When I run train_many_cluster.py
the errors as follows:
Traceback (most recent call last):
File "/home/zxm/Test/fine-grained-segmentation-networks/train/train_many_clustering.py", line 49, in
train_with_clustering_experiment(args.copy())
File "/home/zxm/Test/fine-grained-segmentation-networks/train/train_with_clustering.py", line 537, in train_with_clustering_experiment
save_folder, tmp_folder = generate_name_of_result_folder(args)
File "/home/zxm/Test/fine-grained-segmentation-networks/train/train_with_clustering.py", line 519, in generate_name_of_result_folder
return os.path.join(results_path, result_folder), os.path.join(global_opts['cache_path'], result_folder)
KeyError: 'cache_path'
We hope to get your help!
the link(https://www.dropbox.com/sh/5nvkfzx4nmn4vis/AADseb0UXKcXl4OVWMAJnkHIa?dl=0)is down,can you share with me again? thanks!
Hi! Thanks for sharing your work!
I'd like to know which model should be used to evaluate my own data. I modified cluster_images_in_folder.py
and loaded rc-100clusters.pth
for a try, but the output images are all black. Is there something wrong? Did I miss anything?
Thanks a lot!
Thank you very much for your answer!
I changed the global_opts example.json file,but I still get the same error.
global_opts example.json
{
"result_path": "/home/zxm/Test/fine-grained-segmentation-networks/Models/fgsn",
"cityscapes_path": "/home/zxm/Test/fine-grained-segmentation-networks/Data/cityscapes",
"vistas_path": "/home/zxm/Test/fine-grained-segmentation-networks/Data/mapillary-vistas-dataset_public_v1.2",
"wildash_root_path": "/home/zxm/Test/fine-grained-segmentation-networks/Data/wilddash",
"robotcar_root_path": "/home/zxm/Test/fine-grained-segmentation-networks/Data/Cross-Seasons-Correspondence/ROBOTCAR",
"robotcar_corr_path": "/home/zxm/Test/fine-grained-segmentation-networks/Data/Cross-Seasons-Correspondence/ROBOTCAR/correspondence_data",
"robotcar_im_path": "/home/zxm/Test/fine-grained-segmentation-networks/Data/Cross-Seasons-Correspondence/ROBOTCAR/images",
"cmu_root_path": "/home/zxm/Test/fine-grained-segmentation-networks/Data/Cross-Seasons-Correspondence/CMU",
"cmu_corr_path": "/home/zxm/Test/fine-grained-segmentation-networks/Data/Cross-Seasons-Correspondence/CMU/correspondence_data",
"cmu_im_path": "/home/zxm/Test/fine-grained-segmentation-networks/Data/Cross-Seasons-Correspondence/CMU/images"
}
Thank you!
hello Mans,
I am trying to download the images of CMU datset based on the .sh file as you mentioned in Cross-Season dataset https://www.dropbox.com/sh/n4gt3eazydg8ny0/AAAthUiwz_kYGheSFJCsprtHa?dl=0.
But the website for the dataset http://3dvis.ri.cmu.edu/data-sets/localization/ is not longer available and we can just download the similar one from https://www.visuallocalization.net/datasets/. But the data folder is organised in different ways. Is it possible for you to upload the CMU dataset you used for training? 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.