pretrained model

[kaishijeng]

@seasonSH

Since CASIA dataset is difficult to get, can you provide your pretrained models for performance evaluation?

Thanks
Kaishi

[seasonSH]

I will upload one PFE model trained with this code.

[kaishijeng]

Thanks,

[seasonSH]

I have uploaded more pre-trained models, including a better one which is trained on Ms-ArcFace.

[kaishijeng]

I will download both pretrained models and try them out.

Thanks,

[kaishijeng]

i use PFE_sphere64_msarcface_am and got the following lfw accuracy:

Euclidean (cosine) accuracy: 0.99717 threshold: 1.50062
MLS accuracy: 0.99833 threshold: -2435.03760

Not sure why MLS accuracy threshold is huge negative number.

I have a question of how to use PFE_sphere64_msarcface_am to do face recognition of 2 face images.
I know preprocessing, alignment and network.extract_feature(image) to obtain mu, sigma_sq.
How to use features of mu, sigma_sq to identify whether these 2 images beloging to the same person or not?

Thanks,

[tanpengshi]

I can no longer evaluate the PFE_sphere64_msarcface_am pretrained model for lfw accuracy! Came across the error here:

    157             print('Metagraph file: %s' % meta_file)
    158             print('Checkpoint file: %s' % ckpt_file)
--> 159             saver = tf.compat.v1.train.import_meta_graph(meta_file, clear_devices=False, import_scope=scope)
    160             saver.restore(self.sess, ckpt_file)

ValueError: Node 'gradients/UncertaintyModule/fc_log_sigma_sq/BatchNorm/cond/FusedBatchNorm_1_grad/FusedBatchNormGrad' has an _output_shapes attribute inconsistent with the GraphDef for output #3: Dimension 0 in both shapes must be equal, but are 0 and 512. Shapes are [0] and [512].

Kindly help! Thank you! :)

[chidaksh]

@tanpengshi were you able to find any workaround for the same? Even I'm stuck at the same issue and unable to move forward