This is for end to end decoder for endoscopic imaging with out-of-focus noise. Note that this repository is not the finalized official version (as the paper is in preparation). Most of my research projects were stored in GitHub Enterprise (@github.mit.edu), so I moved this repository here.

• Python 3.6
• Tensorflow 1.15.0
• NumPy
• colorlog
• imageio

• Create a folder under ./data/ for each mouse.
• Create three sub-folders: train/, val/, and test/.
• Create .txt file (like ./data/mouse1/train/label1.txt) for each sub-folder. Each row records: image name, location x, location y, orientation, velocity.

python trainer.py --dataset_path [default: data/mouse1] --batch_size 36 --num_d_conv 6 --num_d_fc 3 --loss_type l1

• The configuration can be found in config.py.

Launch TensorBoard and go to the specified port, you can see different the loss in the scalars tab.