hazelAutumn / ModularBVQA

CVPR2024 ModularBVQA

CVPR2024 Modular Blind Video Quality Assessment

Usage

Prior to commencing, please ensure that you download the relevant datasets and modify the file path accordingly.

Prepare Spatial Rectifier

python3 ModularBVQA-Demo/extract_laplacian/lp_features.py

Prepare Temporal Rectifier

python3 ModularBVQA/extract_motion/extract_SlowFast_feature_others.py

Prepare Base Quality Predictor

python3 ModularBVQA/extract_frame/extract_frame_konvid1k.py

Model

You can download the trained model via Google Drive: https://drive.google.com/file/d/1WP1DhuUgSOusGc3eqLThTSab2-FMBaJA/view?usp=drive_link.

Finetune and evaluate the model

python3 -u ModularBVQA/train_other_modular.py \
--database KoNViD-1k \
--model_name ViTbCLIP_SpatialTemporal_modular_dropout \
--conv_base_lr 1e-5 \
--epochs 30 \
--train_batch_size 16 \
--print_samples 1000 \
--num_workers 16 \
--resize 224 \
--crop_size 224 \
--decay_ratio 0.9 \
--decay_interval 2 \
--loss_type plcc \
--trained_model ModularBVQA-Demo/ckpts_modular/ViTbCLIP_SpatialTemporal_modular_LSVQ.pth

Evaluate the model pretrained on LSVQ

python3 -u ModularBVQA/test_baseline_modular.py \
--database KoNViD-1k \
--train_database LSVQ \
--model_name ViTbCLIP_SpatialTemporal_modular_dropout \
--trained_model ModularBVQA/ckpts_modular/ViTbCLIP_SpatialTemporal_modular_LSVQ.pth \
--num_workers 16 \
--data_path / \
--resize 224 \
--crop_size 224