Closed WatsonWangZh closed 3 years ago
Can you answer my question, I'm trying to reproduce the result? thanks.
Hello, it should be included in the downloaded annotation zip files for GQA or VG8K provided in the README. Did you already download the annotations?
Thank you for your reply. There is a file name "rel_annotations_train.json", is the CSV file the same as JSON file?
Hello, it should be included in the downloaded annotation zip files for GQA or VG8K provided in the README. Did you already download the annotations?
You're welcome. No they are not the same. But why do you need rel_annotation_train.csv? if you are trying to run frequency_based_analysis_of_methods.py on its own, note that this not the correct way to do the evaluation. The evaluation is done through running test_net_rel.py file as mentioned in the README file.
Hi, I have the same problem as OP. Basically when test_net_rel.py is run, specifically line 251:
get_many_medium_few_scores(csv_file, cutoffs, cfg.DATASET, data_dir, ann_dir, syn=True)
it will invokes methods from frequency_based_analysis_of_methods.py which will then requires rel_annotation_train.csv. But as OP mentioned, no rel_annotation_train.csv is included in the provided annotations zip file. Can you provide the csv file? Or can I use the JSON file somehow? Thanks.
Oh another problem that I found is there is no words_synsets.json in the vg8k annotations folder. This will also cause error in test_net_rel.py line 251.
Thank you for providing a thorough explanation of the issue. I now see the problem. I updated the README with the new links for the data folders that contain the needed files. Please check them and confirm with me if you were able to make it work. Thanks.
How to get the 'rel_annotations_train.csv' file? What's format does it have?