To-Do v0.2

[adityaarun1]

• Ensure python3 compatibility
• Fix Tensorboard integration (Crayon does not work for me! ).
• Create separate train and validation split. Currently it uses entire train-val of voc07 for training.
• Improve the baseline (from 66.1% mAP to 69.9% mAP).
• Create separate branch for Fast RCNN. Ensure it works with Selective Search, MCG and Edge Boxes proposals.
• Remove OpenCV display option, integrate them with Tensorboard/Visdom.

[adityaarun1]

The current best performing baseline gives 68.2% mAP. This still needs further improvement.

[adityaarun1]

Tensorboard support is removed. Added Visdom support for visualization of graphs.