Open Catchher opened 4 years ago
halo, any tricks?
halo, any tricks?
No tricks, only increased the batchsize and learning rate to 128 and 1e-1 respectively.
halo, any tricks?
No tricks, only increased the batchsize and learning rate to 128 and 1e-1 respectively.
your cfp-fp performance is quite good~ ! better than my lots of work on it
pre-trained model, this model trained by multi-gpu, when you test it with a single gpu, please remove 'module.' in any variable name. https://pan.baidu.com/s/1jZQ2dp_4YR5TvA4ARkKZ-g code: gh6e
i am currently new in face recognition project, i want to ask what is actual difference between your model and this rep provided? does the model also improve accuracy in face recognition or only improve at face detection?
i am currently new in face recognition project, i want to ask what is actual difference between your model and this rep provided? does the model also improve accuracy in face recognition or only improve at face detection?
No difference to this repo. I advise you to read Arcface and RetinaFace paper. and you will find the answer.
@Catchher Hi, Could you please release the head and optimizer model? I'd like to fine tune on my dataset. Thank you! ☺️
@Catchher would you upload it somewhere else! the baidu is pretty much useless for the rest of the world!
Thanks for sharing your nice work. I have tried your code to reproduce the result with multi-gpu, and got a better performance than your pre-trained model. Following is eval result of ir_se50 model I got after 20 epochs. vgg2_fp - accuray:0.9507999999999999, threshold:1.6540000000000004 agedb_30 - accuray:0.9788333333333334, threshold:1.5299999999999998 calfw - accuray:0.9598333333333333, threshold:1.4600000000000002 cfp_ff - accuray:0.9961428571428572, threshold:1.4459999999999997 cfp_fp - accuray:0.982, threshold:1.59 cplfw - accuray:0.9285, threshold:1.61 lfw - accuray:0.9981666666666665, threshold:1.3960000000000001 I'll upload the model and provide a link later.
What is the value m
Thanks for sharing your nice work. I have tried your code to reproduce the result with multi-gpu, and got a better performance than your pre-trained model. Following is eval result of ir_se50 model I got after 20 epochs.
vgg2_fp - accuray:0.9507999999999999, threshold:1.6540000000000004 agedb_30 - accuray:0.9788333333333334, threshold:1.5299999999999998 calfw - accuray:0.9598333333333333, threshold:1.4600000000000002 cfp_ff - accuray:0.9961428571428572, threshold:1.4459999999999997 cfp_fp - accuray:0.982, threshold:1.59 cplfw - accuray:0.9285, threshold:1.61 lfw - accuray:0.9981666666666665, threshold:1.3960000000000001
I'll upload the model and provide a link later.
pre-trained model, this model trained by multi-gpu, when you test it with a single gpu, please remove 'module.' in any variable name. https://pan.baidu.com/s/1jZQ2dp_4YR5TvA4ARkKZ-g code: gh6e
@Catchher Thanks a lot for sharing weights! I am wondering which data did you use for training. Did you only use "emore" dataset ?
Hi @Catchher, Can you upload pre-train model on google drive or dropbox? Thank you very much!
Hi, I want train this project with multi-GPUs, but meet a problem like "ValueError: some parameters appear in more than one parameter group" in file "Learner.py"(about p47, "self.optimizer = optim.SGD"). Could you tell how to solve this problem or give your multi-gpu train method? Thank you very much!
Hi , what's the different between ir_se50 and original senet50 in Kaiming He 's paper . Any previous work apply imposed ir_se50 ?
@Catchher Thanks for your sharing. I don't know why the acc is 0.6655 and not 0.9981666666666665 when I ran the evaluation code.
@Catchher Thanks for your sharing. I don't know why the acc is 0.6655 and not 0.9981666666666665 when I ran the evaluation code.
Have you found the reason?? I meet the same problem. Look forward to your reply!!
Thanks for sharing your nice work. I have tried your code to reproduce the result with multi-gpu, and got a better performance than your pre-trained model. Following is eval result of ir_se50 model I got after 20 epochs.
vgg2_fp - accuray:0.9507999999999999, threshold:1.6540000000000004 agedb_30 - accuray:0.9788333333333334, threshold:1.5299999999999998 calfw - accuray:0.9598333333333333, threshold:1.4600000000000002 cfp_ff - accuray:0.9961428571428572, threshold:1.4459999999999997 cfp_fp - accuray:0.982, threshold:1.59 cplfw - accuray:0.9285, threshold:1.61 lfw - accuray:0.9981666666666665, threshold:1.3960000000000001
I'll upload the model and provide a link later.