gaopengcuhk / Stable-Pix2Seq

A full-fledged version of Pix2Seq
Apache License 2.0
235 stars 20 forks source link

Performance compared to paper #3

Closed sauradip closed 2 years ago

sauradip commented 2 years ago

Hi ,

Thank you for making this implementation so fast ! How is the difference of this implementation with the reported numbers in the paper ?

gaopengcuhk commented 2 years ago

Please wait a moment

gaopengcuhk commented 2 years ago

Please wait few days. The data loader in stable-pix2seq has been updated. I am retraining stable-pix2seq with 1333 image resolution while old stable-pix2seq uses 512 image resolution by default.

gaopengcuhk commented 2 years ago

By adding new augmentation, stable-pix2seq could achieve 34mAP at 218 epoches

gaopengcuhk commented 2 years ago

256 epoch 36mAP. Beam search will addition improve 1-2 mAP. The estimated mAP is 37-38mAP at 256 epoch.

gaopengcuhk commented 2 years ago

We released the pretrained model at https://github.com/gaopengcuhk/Pretrained-Pix2Seq.