Francis Chen

Computation of SSIM takes so much GPU memory?

1

When computing the ssim of the output of a CNN (batch size 8, image resolution is 192*192), the consumed GPU memory are doubled. It seems that the ssim should require...

Dataset file MD5 checksum problem

1

Hello, I download the Pcam, and compute the MD5 checksum of files by md5sum on Ubuntu 16.04. The relationship between files and MD5 seems promiscuous. E.g. on my computer, the...

Prostate dataset and dataloader

1

Hi Quande, Thanks for sharing this exciting research work! Here I want to learn more details about the implementation of the prostate task. It seems the main scripts (including train_ELCFS.py,...

In the paper, the inference network only contains a single conv layer, and it is conducted with a D-times loop. In this implementation, it seems construct D conv layers with...

Hi, thanks for sharing this project. I am curious about the targets in the forward function. Are the targets as one-hot labels or not? What's the shape of targets, N*C...

yaml configuration of validation and test splits for downstream datasets

2

Hi, thanks for sharing this great project! I am transferring the minigpt-4 to a downstream dataset. The training of downstream dataset is fine, but for the validation/test splits I encountered...

Warning of right-padding for llama_tokenizer.padding_side

1

In MiniGPT4 forward(), the llama_tokenizer is set as padding_side="right", as https://github.com/Vision-CAIR/MiniGPT-4/blob/3bd99950f0ebcbbc7ee7b54aa33f332feeccef09/minigpt4/models/mini_gpt4.py#L177 But when performing llama_model.generate(), this would cause the warning, as follows: `A decoder-only architecture is being used, but right-padding...

Are the items in KL divergence disordered?

1

Hi, In trainer.py, Line 201-Line208, `for i in range(self.model_num): ce_loss = self.loss_ce(outputs[i], labels) kl_loss = 0 for j in range(self.model_num): if i!=j: kl_loss += self.loss_kl(F.log_softmax(outputs[i], dim = 1), F.softmax(Variable(outputs[j]), dim=1))...

Thanks for sharing this great project. Here I have a question on the 2D pixel shuffle. The vit_embeds has the shape of [N, L, C], and it is first reshaped...