search

ViTAE-Transformer / DeepSolo

The official repo for [CVPR'23] "DeepSolo: Let Transformer Decoder with Explicit Points Solo for Text Spotting" & [ArXiv'23] "DeepSolo++: Let Transformer Decoder with Explicit Points Solo for Multilingual Text Spotting"
Other
241 stars 33 forks source link

IC13 finetune Config #15

Closed linxid closed 1 year ago

linxid commented 1 year ago

Can you provide IC13 finetune config file, thx

ymy-k commented 1 year ago

BASE: "../Base_det.yaml"

MODEL: WEIGHTS: "output/R50/150k_tt_mlt_13_15/pretrain/model_final.pth" TRANSFORMER: INFERENCE_TH_TEST: 0.25

DATASETS: TRAIN: ("ic13_train",) TEST: ("ic13_test",)

INPUT: MIN_SIZE_TEST: 1000 MAX_SIZE_TEST: 1824 ROTATE: False

SOLVER: IMS_PER_BATCH: 8 BASE_LR: 1e-5 LR_BACKBONE: 1e-6 WARMUP_ITERS: 0 STEPS: (100000,) # no step MAX_ITER: 1000

linxid commented 1 year ago

Thanks a lot

linxid commented 1 year ago

"ic13_train" and "ic13_test" is "ic13/train_96voc.json" and "ic13/train_37voc.json" in builtin.py:_PREDEFINED_SPLITS_TEXT ?

ymy-k commented 1 year ago

Use "ic13/train_37voc.json". "ic13_test" is not uploaded

linxid commented 1 year ago

Can you upload ic13 test data, we need to evulate model on this dataset. Thx

ymy-k commented 1 year ago

You can refer to SPTS. I remember I downloaded the data and files about ic13 from their repo.