LFFD-A-Light-and-Fast-Face-Detector-for-Edge-Devices icon indicating copy to clipboard operation
LFFD-A-Light-and-Fast-Face-Detector-for-Edge-Devices copied to clipboard

Published 20 hours ago • verified publisher icon YonghaoHe

Pretrained model for validation

Open joyalbin opened this issue 5 years ago • 3 comments

First, glad to see the catching output results with a light model. Thank you for sharing the source code as well.

The pretrained model is not presented at (./saved_model/configuration_10_320_20L_5scales_v2/train_10_320_20L_5scales_v2_iter_1000000.params)

To use the model for my testing, could you please share the pretrained model.

Regards, Albin

joyalbin avatar Nov 14 '19 11:11 joyalbin

@joyalbin The pretrained model does exist there, please check carefully.

YonghaoHe avatar Nov 23 '19 14:11 YonghaoHe

The demo code asks for train_10_320_20L_5scales_v2_iter_1800000.param while in the repository there is only train_10_320_20L_5scales_v2_iter_1000000.params.

Is there a performance difference? Which one are you using?

RoyiAvital avatar Jun 28 '20 09:06 RoyiAvital

@RoyiAvital Just correct the file name in the code to the one name of the saved model. If you want a model that has been trained for 1800000 loops, then you might have to train it yourself. P.S - The code that is there trains for 2000000 loops, you might have to edit that in the config file.

CodexForster avatar Mar 24 '21 06:03 CodexForster