search

cunjian / face_alignment

caffe face alignment 300w
76 stars 32 forks source link

performance #4

Open sunke123 opened 4 years ago

sunke123 commented 4 years ago

Good work.

Could you provide the model size for these models, like Res2Net-50 and so on？ HRNet-W18 gets better performance than provided in this table. You can check this repo. https://github.com/HRNet/HRNet-Facial-Landmark-Detection.

cunjian commented 4 years ago

Hi, thanks for your interest. To clarify, HRNet-Facial-Landmark-Detection uses heatmap-based regression. Here, it is a coordinate-based regression, which pre-train on their ImageNet-based models. Hence, the performance is different. I have yet to integrate the heatmap-based methods.

From: Ke Sun notifications@github.com Sent: Tuesday, March 10, 2020 8:58 AM To: cunjian/face_alignment face_alignment@noreply.github.com Cc: Subscribed subscribed@noreply.github.com Subject: [cunjian/face_alignment] performance (#4)

Good work.

Could you provide the model size for these models, like Res2Net-50 and so on？ HRNet-W18 gets better performance than provided in this table. You can check this repo. https://github.com/HRNet/HRNet-Facial-Landmark-Detection.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/cunjian/face_alignment/issues/4?email_source=notifications&email_token=ADZW57DVOJFQBJDBQSHNNODRGY2PXA5CNFSM4LE6JGN2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IT4ENGQ, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ADZW57C6HLNYYBLFEFUDGS3RGY2PXANCNFSM4LE6JGNQ.