How to train a model from scratch？

mxin262 / Bridging-Text-Spotting

(CVPR 2024) Bridging the Gap Between End-to-End and Two-Step Text Spotting.

https://arxiv.org/pdf/2404.04624.pdf

Other

48 stars 1 forks source link

How to train a model from scratch？ #3

Closed radish512 closed 6 months ago

radish512 commented 7 months ago

Dear author, I noticed that only the pretraining code is provided in the document, like python tools/train_net.py --config-file configs/Bridge/TotalText/R_50_poly.yaml --num-gpus 4 MODEL.WEIGHTS totaltext_final.pth

Is there any code available for training from scratch?

mxin262 commented 6 months ago

This command is for training from scratch. The MODEL.WEIGHTS totaltext_final.pth represents the weight of DPText-DETR or TESTR. The pretrained DiG is presented here

radish512 commented 6 months ago

Do you mean that by removing MODEL.WEIGHTS totaltext_final.pth, it becomes the code for training the model from scratch?

mxin262 commented 6 months ago

Our method is based on the pretrained detector and recognizer. Therefore, you need to load pre-trained detectors and recognizers. If you want to train the detector and recognizer from scratch, you can remove the file MODEL.WEIGHTS totaltext_final.pth, and it will become an end-to-end text spotting system.