search

cleardusk / 3DDFA_V2

The official PyTorch implementation of Towards Fast, Accurate and Stable 3D Dense Face Alignment, ECCV 2020.
MIT License
2.9k stars 514 forks source link

NME metric #20

Open HaoKun-Li opened 4 years ago

HaoKun-Li commented 4 years ago

Hello, thanks for your excellent work!

The NME of different datasets is an important metric. For a fair comparison, may you share the code about how to calculate the NME? Or are there any official code to calculate the NME metrics, and the visibility vector which is shown in "Pose-Invariant 3D Face Alignment(ICCV 2015)" as follow: image image

Looking forward to your reply. Good luck!

cleardusk commented 4 years ago

For the NME part, you can refer to the benchmark of 3DDFA, here. For PIFA, you may refer to the public code in their project page.

shachargluska commented 3 years ago

Can you share the expected NME of the three onnx models (mobilenet, mobilenet0.5, resnet22) on both datasets? I've implemented it myself but I'm not sure I got the correct numbers.

fashionguy commented 3 years ago

@shachargluska could you share your code.

shachargluska commented 3 years ago

I hacked pieces from this repo to the benchmark of 3DDFA Unfortunately I didn't keep it. I do have the results:

mobilenet_v1_1.0_120x120 - 3.68% mobilenet_v1_0.5_120x120 - 3.80% resnet_22_120x120 - 3.67%

Those are nme over aflw2k3d

lewisandJiang commented 3 years ago

I hacked pieces from this repo to the benchmark of 3DDFA Unfortunately I didn't keep it. I do have the results:

mobilenet_v1_1.0_120x120 - 3.68% mobilenet_v1_0.5_120x120 - 3.80% resnet_22_120x120 - 3.67%

Those are nme over aflw2k3d

Could you share the code ? I will appreciate it very much, Sir!

shachargluska commented 3 years ago

I hacked pieces from this repo to the benchmark of 3DDFA Unfortunately I didn't keep it. I do have the results: mobilenet_v1_1.0_120x120 - 3.68% mobilenet_v1_0.5_120x120 - 3.80% resnet_22_120x120 - 3.67% Those are nme over aflw2k3d

Could you share the code ? I will appreciate it very much, Sir!

@lewisandJiang Sorry, but I didn't save this work.

laceyliao commented 2 years ago

I hacked pieces from this repo to the benchmark of 3DDFA Unfortunately I didn't keep it. I do have the results:

mobilenet_v1_1.0_120x120 - 3.68% mobilenet_v1_0.5_120x120 - 3.80% resnet_22_120x120 - 3.67%

Those are nme over aflw2k3d @shachargluska

I have applied benchmark.py3DDFA to 3DDFA_v2 by changing the load model codes, but I got weird results: mobilenet_v1_1.0_120x120 23.654% (aflw20003d) and 22.853%(aflw), should I change the "reconstruct_vertex" code or the code to calculate nme? Hoping for your reply, I will appreciate it very much, Sir!