search

ShawnXYang / TIP-IM

Apache License 2.0
39 stars 5 forks source link

What's the datset that you have used for training the face recognition model? #8

Closed GeorgeCarpenter closed 2 years ago

GeorgeCarpenter commented 2 years ago

Thanks.

ShawnXYang commented 2 years ago

Hi, the trained models are based on the MS-Celeb-1M dataset.

GeorgeCarpenter commented 2 years ago

Could you tell us the size of the input images for these face recognition models?

GeorgeCarpenter commented 2 years ago

Could you tell us the size of the input images for these face recognition models?

The paper write:

In experiments, we first use MTCNN [53] to detect faces in the image, then align the images and crop them to 112 × 112, meaning that the identity masks are executed only in the face area.

However, we have tried the pre-trained models you have provided. We found that the size of the input images of ArcFace is (112, 96).

Could you release the correct model you have used whose size of the input images is (112, 112)?

ShawnXYang commented 2 years ago

Please refer to https://github.com/ShawnXYang/TIP-IM/blob/master/get_model.py#L19 . We further adaptively resize these images because the input requirement of each pre-trained model is different.