d239: Translation-aware Fully Convolutional Instance Segmentation, Winner of Common objects in context (COCO2016)

Microsoft COCO: Common Objects in Context [Paper]: https://arxiv.org/abs/1405.0312 PDF: https://arxiv.org/pdf/1405.0312v3

Repository and source code for TA-FCN, the winning entry of COCO segmentation challenge 2016: https://github.com/daijifeng001/TA-FCN

screen-shot-2016-10-10-at-10-21-08

Slides: https://onedrive.live.com/?authkey=%21AEYOyOirjIutSVk&cid=F371D9563727B96F&id=F371D9563727B96F%2197213&parId=F371D9563727B96F%2192189&o=OneUp

Visual results on the first 5k images from COCO test set: https://onedrive.live.com/?authkey=%21ABB_CV2zvCEoNK0&id=F371D9563727B96F%2192190&cid=F371D9563727B96F

About COCO: http://mscoco.org/

Dataset: http://mscoco.org/dataset