examples

Examples

0. Preparation

Please download and extract your dataset to ./data, e.g. data/StanfordDogs

Dataset list:

• data/StanfordDogs: http://vision.stanford.edu/aditya86/ImageNetDogs/
• data/StanfordCars: https://ai.stanford.edu/~jkrause/cars/car_dataset.html
• data/CUB200: http://www.vision.caltech.edu/visipedia/CUB-200.html
• data/FGVC-Aircraft: http://www.robots.ox.ac.uk/~vgg/data/fgvc-aircraft/
• data/NYUv2: https://github.com/VainF/nyuv2-python-toolkit

Visualize training with Tensorboard

tensorboard --logdir=run

1. Scratch Training

Finegraind Classification

python finegraind.py --dataset stanford_dogs

Semantic Segmentation on NYUv2

python nyuv2_seg.py

Monocular Depth Estimation on NYUv2

python nyuv2_depth.py

2. Knowledge Amalgamation

cd knowledge_amalgamation

Task Branching

python joint_scene_parsing_task_branching.py --seg_ckpt PATH_TO_YOUT_SEG_MODEL --depth_ckpt PATH_TO_YOUT_DEPTH_MODEL

Common Feature Learning

python comprehensive_classification_common_feature_learning.py --car_ckpt PATH_TO_YOUR_CAR_CLASSIFIER --aircraft_ckpt PATH_TO_YOUT_AIRCRAFT_CLASSIFIER

Layerwise Amalgamation

python comprehensive_classification_layerwise_amalgamation.py --car_ckpt PATH_TO_YOUR_CAR_CLASSIFIER --aircraft_ckpt PATH_TO_YOUT_AIRCRAFT_CLASSIFIER