Thank you for this modern toolchain for face recogntion training.
There is a small issue in the usage of the insightface models here:
|
def from_config(cls, config): |
For the resnet100 the output_dim config entry is missing, for 18 and 50 the parameter has a wrong name. It doesn't expect output_dim=config.output_dim but num_features=config.output_dim. The input_size parameter is not supported by the model at all (but would be helpful).
Also Resnet200 is missing, but that's not really a bug.
Thank you for this modern toolchain for face recogntion training.
There is a small issue in the usage of the insightface models here:
CVLface/cvlface/research/recognition/code/run_v1/models/iresnet_insightface/__init__.py
Line 25 in 308142a
For the resnet100 the output_dim config entry is missing, for 18 and 50 the parameter has a wrong name. It doesn't expect
output_dim=config.output_dimbutnum_features=config.output_dim. The input_size parameter is not supported by the model at all (but would be helpful).Also Resnet200 is missing, but that's not really a bug.