Skip to content

World Knowledge Distillation in NN using Masked Target Representations

License

Notifications You must be signed in to change notification settings

thunderock/BlindNet

Repository files navigation

BlindNet: Scene Based Knowledge Distillation

UNET: Pixel level prediction for each bounding box

Dataset structure:

-datasets/coco/annotations/

-datasets/coco/val2017/

-datasets/coco/train2017/

python src/train_unet.py

Simple CNN: predict only the masked object

python src/train_single_maskcnn.py

EXPERIMENT 3 (Pixel wise class id prediction)

First download dataset from here.

to run the experiment, run the following command:

python driver.py

EXPERIMENT 4 (Multilabel classification of classes in a scene)

First download dataset from here.

to run the experiment, run the following command:

python driver_multilabel.py

About

World Knowledge Distillation in NN using Masked Target Representations

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published