Skip to content

Latest commit

 

History

History
39 lines (25 loc) · 822 Bytes

README.md

File metadata and controls

39 lines (25 loc) · 822 Bytes

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