search

lkevinzc / dance

Codes for "DANCE: A Deep Attentive Contour Model for Efficient Instance Segmentation", WACV2021
67 stars 13 forks source link

which part in your code re-implemented DeepSnake with FCOS as the object detector? #1

Closed cccccccccy closed 3 years ago

cccccccccy commented 3 years ago

I have read your paper, and it mentioned you re-implemented DeepSnake with FCOSas the object detector for fair comparison. So can you tell me which part re-implemented DeepSnake with FCOS in your code? Thanks !

lkevinzc commented 3 years ago

Hi! Thanks for your interest!

https://github.com/lkevinzc/dance/blob/master/core/modeling/edge_snake/snake_head.py#L617

Starting from here, it uses a snake module to first generate extreme points and then create the octagon, afterwards do mask regression.

cccccccccy commented 3 years ago

Hi! Thanks for your excellent work! I still have another question. FCOS generates bbox_prediction with HW4 dimensions, and in my opinion we need to select right bboxes as the deformation's input. Is my understanding correct? And if it is correct, how to select right bboxes and which part of your code implements this work? Thanks for your reply!

lkevinzc commented 3 years ago

@cccccccccy The NMS process should select those high quality object boxes for you. The deformation is based on those selected boxes. Here is where the FCOS does NMS as post-processing: https://github.com/lkevinzc/dance/blob/57ffc777c52024c351293fe39ba217ae7f1fa24f/core/modeling/fcose/fcose_outputs.py#L528

cccccccccy commented 3 years ago

Hi! , One error occurred，when I used coco2017 dataset for training. The command is python train_net.py --num-gpus 1 --config-file configs/Dance_R_50_3x.yaml And the error is FileNotFoundError: [Errno 2] No such file or directory: 'datasets/coco/edge_train2017' Does this still require some preprocessing files for training？

lkevinzc commented 3 years ago

@cccccccccy Hi, yes. You need to use this script to prepare the training data first.