search

vitoralbiero / img2pose

The official PyTorch implementation of img2pose: Face Alignment and Detection via 6DoF, Face Pose Estimation - CVPR 2021
Other
588 stars 109 forks source link

Question about Fine tuning Model with 300W-LP #68

Closed RookieDookie123 closed 2 years ago

RookieDookie123 commented 2 years ago

Hi @vitoralbiero ,

I have read the Section 4.2 stated in the paper but I still can't understand how should I fine-tune my model with the dataset after training it with WIDER FACE. Could you tell me the steps on fine-tuning the model ?

Thank you in advance. Your helps are much appreciated.

vitoralbiero commented 2 years ago

Hello @RookieDookie123,

Thanks for your interest in our paper. After training on WIDERFACE, load the trained weights, in models.py change rpn_batch_size_per_image to 2 (proposals), and box_detections_per_img to 4 (head samples). Then, fined tune with fixed learning rate (0.001) for 2 epochs.

I hope this helps.