prajwalsingh / EEG2Image

EEG2IMAGE: Image Reconstruction from EEG Brain Signals. [ICASSP 2023]
MIT License
61 stars 9 forks source link

Question about GAN Training #16

Closed binnnnnny closed 5 months ago

binnnnnny commented 5 months ago

Hi, thanks for your great work ! I have some questions while applying my dataset.

  1. When learning GAN for train.pyin test_batch, the image is a training dataset but EEG signal is test data. Should I go through the inference process once more with test_generation.py after learning at train.py ?

  2. I'm trying to get the Inception Score of the paper, but the results are somehow different from the values in the paper. I got 8.04 Inception Score.

1) Prepare your dataset, LSTM ckpt, GAN ckpt
2) Run test_generation.py
3) Run inceptionscore.py

Can you tell me what the problem is?

Thanks!

prajwalsingh commented 5 months ago

Hi @binnnnnny , thank you for reading our work.

  1. The reason why test images are similar to train images is because of the way the dataset is collected. For more details, refer to this issue [prajwalsingh/EEG2Image/issues/4].

  2. Please refer to this [prajwalsingh/EEG2Image/issues/2] regarding the inception score clarification.