A Comparison Based Study On Depth Estimation of Monocular Endoscopic Images using Self-supervised Learning Methods
Training dataset root : https://drive.google.com/drive/folders/122y00jCDJswnLwgNwyFhE3cdyg_EzX6j?usp=sharing
Training of our implemented model from scratch:
!python train.py --id_range 2 --input_downsampling 4.0 --network_downsampling 64 --adjacent_range 5 30 - -input_size 256 320 --batch_size 2 --num_workers 2 --num_pre_workers 2 --validation_interval 1 --display_interval 50 --dcl_weight 5.0 --sfl_weight 20.0 --max_lr 1.0e-3 --min_lr 1.0e-4 --inlier_percentage 0.99 --visibility_overlap 30 --training_patient_id 1 --testing_patient_id 1 --validation_patient_id 1 --number_epoch 51 --num_iter 100 --training_result_root "./Pre_trained_models" --training_data_root "./training_data_root"
Training of our implemented model by loading pre-trained model:
!python train.py --id_range 2 --input_downsampling 4.0 --network_downsampling 64 --adjacent_range 5 30 --input_size 256 320 --batch_size 2 --num_workers 2 --num_pre_workers 2 --validation_interval 1 --display_interval 50 --dcl_weight 5.0 --sfl_weight 20.0 --max_lr 1.0e-3 --min_lr 1.0e-4 --inlier_percentage 0.99 --visibility_overlap 30 --training_patient_id 1 --testing_patient_id 1 --validation_patient_id 1 --number_epoch 51 --num_iter 100 --training_result_root "./Pre_trained_models" --training_data_root "./training_data_root" --load_trained_model --trained_model_path "./Pre_trained_models/depth_estimation_train_run_8_3_18_8_test_id_[1]/"checkpoint_model_epoch_51_validation_0.08002564724948671.pt
TensorBoard Visualisation of results:
!pip install tensorboardX %load_ext tensorboard %tensorboard --logdir './Pre_trained_models/depth_estimation_train_run_7_12_16_16_test_id_[1]'
Testing with pre-trained model with natural images:
!python test_simple.py --image_path assets/test_image.jpg --model_name mono+stereo_640x192
Refernce repositories:
Link for FC Densenet model genaration for training: https://github.com/SimJeg/FC-DenseNet
Link for cyclical learning rate: https://github.com/bckenstler/CLR
Link for single image depth estimation in endoscopy: https://github.com/lppllppl920/EndoscopyDepthEstimation-Pytorch
Link for single image depth estimation in natural data set: https://github.com/nianticlabs/monodepth2