search

MengyangPu / EDTER

EDTER: Edge Detection with Transformer, in CVPR 2022
MIT License
277 stars 35 forks source link

Fine Tuning on our custom dataset #91

Open Arnab1181412 opened 9 months ago

Arnab1181412 commented 9 months ago

Hi I have image pairs for my own data and I want to fine tune using this dataset. There are around 500 images of 255 X 255 size. How do I do it? Thanks in advance

MengyangPu commented 9 months ago

@Arnab1181412 The process is the same as the training process: 1 Data augment, and generate a train_pair.txt of your data. 2 Modify the path in the configuration file. 3 Adjust parameters, e.g., learning rate, and batch size.

Arnab1181412 commented 9 months ago

Thank you for your help. Just had one question: should I change the path at this line https://github.com/MengyangPu/EDTER/blob/3fe76f3d938206ef9dc8b857a9767b8cd3d28fc7/configs/bsds/EDTER_BIMLA_320x320_80k_bsds_bs_8.py#L3C66-L3C66. ?

Regards

On Tue, Jan 16, 2024 at 8:36 AM Mengyang Pu @.***> wrote:

The process is the same as the training process https://github.com/MengyangPu/EDTER?tab=readme-ov-file#2-training: 1 Data augment, and generate a train_pair.txt https://github.com/MengyangPu/EDTER/blob/main/data/BSDS/ImageSets/train_pair.txt of your data. 2 Modify the path in the configuration file https://github.com/MengyangPu/EDTER/blob/main/configs/bsds/EDTER_BIMLA_320x320_80k_bsds_bs_8.py . 3 Adjust parameters, e.g., learning rate, and batch size.

— Reply to this email directly, view it on GitHub https://github.com/MengyangPu/EDTER/issues/91#issuecomment-1893003831, or unsubscribe https://github.com/notifications/unsubscribe-auth/ASBBP455YEFY6KNQ2RPCEULYOXVEHAVCNFSM6AAAAABB3UEDN2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQOJTGAYDGOBTGE . You are receiving this because you authored the thread.Message ID: @.***>

MengyangPu commented 9 months ago

Yes, and you could create your dataset.py based on configs/base/datasets/bsds.py.

Arnab1181412 commented 9 months ago

Thank you for your help. Let me try this. I will get back to you if I have any questions.

Regards

On Tue, Jan 16, 2024 at 6:29 PM Mengyang Pu @.***> wrote:

Yes, and you could create your dataset.py based on configs/base /datasets/bsds.py https://github.com/MengyangPu/EDTER/blob/3fe76f3d938206ef9dc8b857a9767b8cd3d28fc7/configs/_base_/datasets/bsds.py .

— Reply to this email directly, view it on GitHub https://github.com/MengyangPu/EDTER/issues/91#issuecomment-1893698520, or unsubscribe https://github.com/notifications/unsubscribe-auth/ASBBP4ZHWUQMJFLGVAUHRA3YOZ2U3AVCNFSM6AAAAABB3UEDN2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQOJTGY4TQNJSGA . You are receiving this because you were mentioned.Message ID: @.***>

Arnab1181412 commented 9 months ago

Hi I tried following the steps you told me but it didn't work. I did it on a kaggle notebook and the linux commands are not working. That is the first training process 2.1.

On Wed, Jan 17, 2024 at 12:54 AM Arnab Samanta < @.***> wrote:

Thank you for your help. Let me try this. I will get back to you if I have any questions.

Regards

On Tue, Jan 16, 2024 at 6:29 PM Mengyang Pu @.***> wrote:

Yes, and you could create your dataset.py based on configs/base /datasets/bsds.py https://github.com/MengyangPu/EDTER/blob/3fe76f3d938206ef9dc8b857a9767b8cd3d28fc7/configs/_base_/datasets/bsds.py .

— Reply to this email directly, view it on GitHub https://github.com/MengyangPu/EDTER/issues/91#issuecomment-1893698520, or unsubscribe https://github.com/notifications/unsubscribe-auth/ASBBP4ZHWUQMJFLGVAUHRA3YOZ2U3AVCNFSM6AAAAABB3UEDN2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTQOJTGY4TQNJSGA . You are receiving this because you were mentioned.Message ID: @.***>