Train with lmdb dataset

wenwenyu / MASTER-pytorch

Code for the paper "MASTER: Multi-Aspect Non-local Network for Scene Text Recognition" (Pattern Recognition 2021)

https://arxiv.org/abs/1910.02562

MIT License

278 stars 52 forks source link

Train with lmdb dataset #1

Open CuongNN218 opened 3 years ago

CuongNN218 commented 3 years ago

I want to train Master with LMDB dataset but I don't see any args related to image size. How can I train with fixed image size ?

CuongNN218 commented 3 years ago

@wenwenyu and how can I train with only one GPU

wenwenyu commented 3 years ago

Hi, LmdbDataset has added img_w and img_h args. Training with one GPU can use debug model. e.g. python train.py -c configs/config.json -d 1 -dist false, and -d specify which one gpu will be used.

CuongNN218 commented 3 years ago

how can I create a lmdb dataset to work with it. Maybe your key in lmdb file is quite different from that of deep-text-recognition-benchmark @wenwenyu

wenwenyu commented 3 years ago

Yes, the key in the lmdb file is different from that of deep-text-recognition-benchmark. Two ways to deal with this difference.

You can modify the key to adapt your lmdb dataset, including nSamples, image_key, and label_key
Refer to the example of creating lmdb dataset to satisfy our LmdbDataset

qyfff commented 1 year ago

我有3个gpu在一个机器上怎么训练