lizhaoliu-lec/denet

This is the official repo for Dynamic Extension Nets for Few-shot Semantic Segmentation (ACM Multimedia 20).

denet's Issues

COCO20i dataset

Is there a problem with the coco.py file? I use the function run_WrappedCOCOStuff20i to download the COCO dataset, but when it comes to the processing of fold, it always reports an error. Why?Sorry to bother you.

where is the paper?

about dateset in config file

after run the function run_VOCSBDSegmentation5i_from_scratch() and copy the content in train_config_denet.template to train_config.yml, when I use "python -u train.py", the error is can't find "WrappedVOCSBDSegmentation5i" which in the config file, but the dataset VOC didn't contain the "WrappedVOCSBDSegmentation5i" after run the function run_VOCSBDSegmentation5i_from_scratch(), so where should I find the "WrappedVOCSBDSegmentation5i" folder? I tried to run other test functions in voc_sbd.py, but I didn't get what I want.

Pretrained models

Would you mind releasing the pretrained models on COCO-20i?

RuntimeError: File not found or corrupted.

Total message:
python voc_sbd.py
Using downloaded and verified file: ./PASCAL-5i/VOCtrainval_11-May-2012.tar
Extracting ./PASCAL-5i/VOCtrainval_11-May-2012.tar to ./PASCAL-5i
Downloading https://www2.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/semantic_contours/benchmark.tgz to ./PASCAL-5i/benchmark.tgz
1419540480it [09:34, 2468838.86it/s]
Extracting ./PASCAL-5i/benchmark.tgz to ./PASCAL-5i
Downloading https://www.cs.cornell.edu/~bharathh/ to ./PASCAL-5i/train_noval.txt
21504it [00:00, 77987.82it/s]
Traceback (most recent call last):
File "voc_sbd.py", line 917, in
run_VOCSBDSegmentation5i_from_scratch()
File "voc_sbd.py", line 682, in run_VOCSBDSegmentation5i_from_scratch
voc = VOCSBDSegmentation5i(roots=roots_path,
File "voc_sbd.py", line 282, in init
super().init(*args, **kwargs)
File "voc_sbd.py", line 89, in init
self.sbd = SBDataset(root=roots or root_sbd,
File "/home/miniconda3/envs/msa/lib/python3.8/site-packages/torchvision/datasets/sbd.py", line 83, in init
download_url(self.voc_train_url, sbd_root, self.voc_split_filename, self.voc_split_md5)
File "/home/miniconda3/envs/msa/lib/python3.8/site-packages/torchvision/datasets/utils.py", line 152, in download_url
raise RuntimeError("File not found or corrupted.")
RuntimeError: File not found or corrupted.

What should I do?

Cannot download PASCAL VOC 5i from your link.

Could you please share a new link for downloading this dataset? Thanks.

Drive links are not shared?

The two links on the main page both give a 404

Evaluation Problem

Thanks for releasing the code. It is an impressive work. And when I look into the whole pipeline, it is a little confused that the mIOU seems averaged on all class? Do you change the maximum_num_classes during testing?

about dataset in config files

after run the function run_VOCSBDSegmentation5i_from_scratch() and copy the content in train_config_denet.template to train_config.yml, when I use "python -u train.py", the error is can't find "WrappedVOCSBDSegmentation5i" which in the config file, but the dataset VOC didn't contain the "WrappedVOCSBDSegmentation5i" after run the function run_VOCSBDSegmentation5i_from_scratch(), so where should I find the "WrappedVOCSBDSegmentation5i" folder? I tried to run other test functions in voc_sbd.py, but I didn't get what I want.