search

pkuCactus / BDCN

The code for the CVPR2019 paper Bi-Directional Cascade Network for Perceptual Edge Detection
MIT License
341 stars 71 forks source link

About the data augmentation of PASCAL Context dataset? #32

Open yuzhegao opened 4 years ago

yuzhegao commented 4 years ago

Hi @pkuCactus From your paper, I have knew that using the PASCAL Context dataset as additional training dataset can help improve the performance( ODS 0.806 -> 0.820)，so I wonder some training detail. Could your please tell me whether you augment the Pascal Context dataset while training?

pkuCactus commented 4 years ago

Yes, we did augment with Pascal Context while training.

ShreyasSkandanS commented 3 years ago

@pkuCactus what information did you use for ground truth edges in Pascal Context, the semantic class boundaries? Do you have the data for this?

I'm trying to train your network with the HED-BSDS dataset provided but haven't been able to achieve these accuracy values after 40k iterations.

pkuCactus commented 3 years ago

Yes, we indeed use the semantic class boundaries of Pascal Context to augment the training data. Moreover, we also do augmentation on the pascal context to further increase the training data. And then combine the augmented bsds and pascal context for training. However, I have graduated and the data are stored in the school server, I'm not available now. You can download the data from the link and do the augmentation by yourself.

ForawardStar commented 3 years ago

@pkuCactus Where is the link? And how to do augmentation?

liuzhidemaomao commented 2 years ago

@pkuCactus Hello~! BDCN is a nice job! Can you tell me what augmentations do you have applied for Pascal Context?