mean-teacher
mean-teacher copied to clipboard
CuriousAI
A state-of-the-art semi-supervised method for image recognition
I trained the ResNet architecture (cifar_shakeshake26 in Pytorch version) on cifar-10 dataset with 1000 unlabeled images and 44000 labeled images (the resting 5000 images are used for validation) for about...
Hello! There is one function that I don't understand in /mean_teacher/utils.py `def export(fn): mod = sys.modules[fn.__module__] if hasattr(mod, '__all__'): mod.__all__.append(fn.__name__) else: mod.__all__ = [fn.__name__] return fn` Look forward for your...
python 3.6 cuda9.0 pytorch0.4
Hi, i am wonder about this loss: ` class_loss = class_criterion(class_logit, target_var) / minibatch_size ` since this loss ignore some samples(no_lable), why here still use the minibatch_size not the labeled_size?
Thanks for your codes. I had to admit it's a wonderful strategy. However, when I use this package on the action recognition dataset Stanford40, I encounter the loss explosion problem,...
Hi, the code will be update for a new version of pytorch? I'm trying to do this by my own, but I'm new to pytorch and finding some issues. After...
mean-teacher has not [initialized](https://github.com/CuriousAI/mean-teacher/blob/master/tensorflow/mean_teacher/weight_norm.py#L41) b as given in Original [weight_norm paper](https://arxiv.org/pdf/1602.07868.pdf) and its [implementation](https://github.com/openai/pixel-cnn/blob/fc86dbce1d508fa79f8e9a7d1942d229249a5366/pixel_cnn_pp/nn.py) that you have [referenced](https://github.com/CuriousAI/mean-teacher/blob/master/tensorflow/mean_teacher/weight_norm.py#L13). Is it intentional?
I am talking about `self.mean_cons_cost_mt, self.cons_costs_mt = consistency_costs( self.cons_logits_1, self.class_logits_ema, self.cons_coefficient, consistency_mask, self.hyper['consistency_trust'])` What is the difference between using 'self.class_logits_ema' and using 'self.cons_logits_ema' as consistency targets?
Great paper! Tensorflow documentation says the EMA variables are created with (trainable=False) and added to the GraphKeys.ALL_VARIABLES collection. Now as they are not trainable they wont have the gradient applied...
File "./main.py", line 166, in main train(train_loader, train_loader_len, model, ema_model, ema_model, optimizer, epoch, training_lo File "./main.py", line 492, in train assert not (np.isnan(loss.data[0]) or loss.data[0] > 1e5), 'Loss explosion: {}'.format(loss.data...
