akhilgakhar / wsol_model_selection

Pytorch code for: Realistic Model Selection for Weakly Supervised Object Localization

Citation:

@InProceedings{murtaza_2024_wsol_model_selection,
  title   = {Realistic Model Selection for Weakly Supervised Object Localization},
  author  = {S. Murtaza and S. Belharbi and M. Pedersoli and E. Granger},
  journal = {CoRR},
  year    = {2024}
}

Issues:

Please create a github issue.

Requirements

See full requirements at ./dependencies/requirements.txt

• Python
• Pytorch
• torchvision
• Full dependencies
• Build and install CRF:
  • Install Swig
  • CRF (not used in this work, but it is part of the code.)

cdir=$(pwd)
cd dlib/crf/crfwrapper/bilateralfilter
swig -python -c++ bilateralfilter.i
python setup.py install
cd $cdir
cd dlib/crf/crfwrapper/colorbilateralfilter
swig -python -c++ colorbilateralfilter.i
python setup.py install

Pseudo-Bboxes Annotations:

• For validation, bounding box annotations can be found at:
• Bounding boxes annotation for validation can be used it one run that we produced is available at:
  • Bboxes annotations for ILSVRC: ./folds/wsol-done-right-pseduo-splits/metadata_generated_by_selective_search_bboxs/ILSVRC/val
  • Bboxes annotations for CUB: ./folds/wsol-done-right-pseduo-splits/metadata_generated_by_selective_search_bboxs/CUB/val
• In our code, you can simultaneously input the names and paths of various bounding boxes by utilizing the parameter --metadata_roots_pseduo_boxs_valset (refer to the provided example). The code will perform evaluation and checkpointing for each bounding box set.

Download datasets:

See folds/wsol-done-right-splits/dataset-scripts. For more details, see wsol-done-right repo.

You can use these scripts to download the datasets: cmds. Use the script _video_ds_ytov2_2.py to reformat YTOv2.2.

Once you download the datasets, you need to adjust the paths in get_root_wsol_dataset().

Model Specific Hyperparamters and their range

Method	Hyperparameter	Sampling Distribution	Range
Common HPs	LR, WD, Gamma	LogUniform	[10^-5,10^0]
Common HPs	Step Size	Uniform	CUB: [5-45], ILSVRC: [2-9]
CAM, TS-CAM, SCM, NL-CCAM	Common HPs	-	-
HaS	Drop Rate, Drop Area	Uniform	[0,1]
ACoL	Erasing Threshold	Uniform	[0,1]
ADL	Drop Rate, Erasing Threshold	Uniform	[0,1]
SAT	Area Threshold	Uniform	[0,1]

Run code:

To present the WSOL baselines with CAM over CUB using ResNet50, and for other methods by simply replacing the method name and indicating the model-specific parameter as detailed above:

cudaid=0  # cudaid=$1
export CUDA_VISIBLE_DEVICES=$cudaid
python main_wsol.py \
    --task STD_CL \
    --encoder_name resnet50 \
    --arch STDClassifier \
    --opt__name_optimizer sgd \
    --batch_size 32 \
    --max_epochs 50 \
    --freeze_cl False \
    --support_background True \
    --method CAM \
    --spatial_pooling WGAP \
    --dataset CUB \
    --box_v2_metric False \
    --cudaid 0 \
    --debug_subfolder None \
    --cam_curve_interval 0.001 \
    --exp_id with_gt_and_pseduo_metadata_id_0 \
    --num_workers 2 \
    --opt__lr 0.0001 \
    --opt__weight_decay 1e-05 \
    --opt__step_size 5 \
    --opt__gamma 0.1 \
    --metadata_roots_pseduo_boxs_valset \{\"metadata_by_clip\":\ \"folds/wsol-done-right-pseduo-splits/metadata_generated_by_clip_maps\",\ \"metadata_by_rpn\":\ \"folds/wsol-done-right-pseduo-splits/metadata_generated_by_rpn_bboxs\",\ \"metadata_by_ss\":\ \"folds/wsol-done-right-pseduo-splits/metadata_generated_by_selective_search_bboxs\"\}