GarField
GarField
Hi, zhang: I am very interested in implemented about dynamic task coefficient in 68 landmark detection?can you give some details with it ? if some codes, it would be better!...
you could add the code in #compress weights module: `#compress weights if len(net.params[name]) > 1 : weights = net.params[name][0].data #don't compress bais bais = net.params[name][1].data else: weights = net.params[name][0].data bais...
> https://github.com/deepinsight/insightface/wiki/Model-Zoo Thank you for your work!!
Voc data contains lots of clean images, which could be used to construct unpaired clean and noisy data easily. In fact, you could use any other dataset to train your...
@Zruto you should firstly train the encoder and generator to produce the model file, and then the '.pt' file is used for testing.
you could retrain the model with your datasets easily, including BSD, SET or Mayo datasets, which could need one-day time to get the final model.
Adding the KL loss for latent codes is only used to validate the effects of loss functions in my exps, e.g., jointing gan loss and the KL loss, but the...
For LDCT denoising, you could use the unit_noise2clear-bn.yaml in configs. In fact, there is few differences in config file for different task training.
It could be viewed as a regularization trick to improve the robustness of the generator.
thanks for your quick reply, this is a nice work, but it is truly for normalized flow models that cost more time to get good results than GANs and VAEs.