search

altndrr / vicss

Code implementation of our paper: Vocabulary-free Image Classification and Semantic Segmentation
MIT License
4 stars 1 forks source link

Excuse me #1

Open ZhaoHuiii opened 4 months ago

ZhaoHuiii commented 4 months ago

Hello, I would like to ask why an error occurs when I change the "experiment" to "method/dense_cased" in the train.yaml file. Is there anything else that needs to be modified? Thank you.

altndrr commented 3 months ago

Hi, changing the experiment reference should be sufficient to have the correct configuration. Are you experiencing a particular issue?

ZhaoHuiii commented 3 months ago

Thank you for your reply. When I modify the experiment to method/dense_cased and the data to ade20k_150, and run train.py again, the code prompts me with the following error: RuntimeError: The size of tensor a (82) must match the size of tensor b (257) at non-singleton dimension 1 Can you tell me which configurations I should modify in order to successfully run the dense_cased experiment? Thank you.

altndrr commented 3 months ago

Can you be more specific with the command you pass to the code? And what is the complete error message? Have you tried other datasets? Did you follow the README to setup the environment?

From a first inspection, the error does not seem related to a wrong input configuration, but I cannot confirm it if I don't get the full output.

Thank you

ZhaoHuiii commented 3 months ago

I pass the command "python src/train.py data=ade20k_150 experiment=method/dense_cased" to the code. If I want to construct my own dataset (images and their descriptions), what should I do? Thank you very much!

altndrr commented 3 months ago

The command should be correct. The issue could be due to the data preprocessing. Maybe the dataset preparation was interrupted and now you have only part of the dataset.

What are the specifics of your dataset? You can take inspiration from the dataset implementations in the src/data submodule to understand how I download, process, and load different datasets. If you work with segmentation datasets, ADE20k, PASCAL Context, and PascalVOC should be good references.