Codes for Language-Conditioned Feature Pyramids for Visual Selection
- some utilities (e.g. dataset converters)
- experiment codes
These files are common except for the configration.
- attribute_selector_onecommon_exp_n.py
- lcfp_selector_onecommon_exp_n.py
- lcfp_selector_exp_n.py
Download related files and locate them in the repository.
- data/guesswhat.train.jsonl
- data/guesswhat.valid.jsonl
- data/guesswhat.test.jsonl
- data/img/train2014
- data/img/val2014
- data/img/test2014
$ python src/lcfp_selector_exp_0.py
The model directory will be created in models directory.
Weights and validation history will be stored in the directory.
$ python src/eval_model_guesswhat.py src/lcfp_selector_exp_0.py
It choose and evaluate a weight with the best validation based on the history.
It output the result to the standard output.
Download related files and locate them in the repository.
- onecommon_data/train.txt
- onecommon_data/train_success_only.txt
- onecommon_data/train_uncorrelated.txt
- onecommon_data/valid.txt
- onecommon_data/valid_success_only.txt
- onecommon_data/valid_uncorrelated.txt
- onecommon_data/test.txt
- onecommon_data/test_success_only.txt
- onecommon_data/test_uncorrelated.txt
Run convert_onecommon_data.py script to get converted files and images.
$ cd <repository root>
$ python convert_onecommon_data.py -i onecommon_data
Converted directory will be created in onecommon_data directory.
$ python src/lcfp_selector_onecommon_exp_0.py
$ python src/eval_model_onecommon.py src/lcfp_selector_onecommon_exp_0.py