From a18ce740da29f74915635e600aad1b1678a6d2bc Mon Sep 17 00:00:00 2001
From: Guillaume Bellec <guillweb@gmail.com>
Date: Thu, 25 Nov 2021 18:08:34 +0100
Subject: [PATCH] Update README.md

---
 README.md | 27 ++++++++++-----------------
 1 file changed, 10 insertions(+), 17 deletions(-)

diff --git a/README.md b/README.md
index ef95ed9..3539f75 100644
--- a/README.md
+++ b/README.md
@@ -23,29 +23,22 @@ python train.py --SEED=0 --ground_truth=no  --GPU=yes --maxiter=800
 ```
 
 
-`src.parser.py` specifies arguments used in the experiment.
-
-
-#### Network architecture
-
-Overall, the network `src.buildnet.py` consists of two parts. One is the CNN `conv2`that pre-processes the video non-linearly. And the other is stochastic recurrent neural network `RSNN` that contains one layer of N GLM neurons that generates spikes.
+The default parameters defined in `src.parser.py` should match the parameters used in the paper.
 
 
 #### Dataset
 
-`DataLoader` in the `src.utils.py` handles preprocessing and loading data.
-
-To run the code, one will need to prepare (saved under folder `data_dir` specified in `src.parser.py`):
-One video data for training, saved in the form of numpy array with size: #pixel_y, #pixel_x, #frame; 
-One spike train recordings for training, saved in the form of numpy array with size: #repetition, #frame, #neuron;
-And similarly, one video data + one spike train recordings for validating/ testing.
-
-You can either use your own dataset, or download the dataset we used (V1-dataset) from https://crcns.org/data-sets/vc/pvc-11/about.
-If you want to use V1-dataset, please follow the instructions of the website and refer to the pre-processing steps described in our paper.
+To reproduce the results on the V1-dataset, download it at: https://crcns.org/data-sets/vc/pvc-11/about.
+Please follow the instructions of the website and refer to the pre-processing steps described in the Appendix A of our paper.
 
-If you are testing a different dataset, you may also want to tailor the CNN that pre-processes the video according to the input.
-Once a different video or CNN is used, one have to specify the `input_latent_size`, it's the size of the flattened output of CNN, also the input size of the recurrent network.
+The processed data should be saved in the folder `data_dir` specified in `src.parser.py`.
+The video data should be saved as a numpy array of shape: ( #pixel_y, #pixel_x, #frame ).
+The spike train aligned to the video should be saved in the form of numpy array of shape: ( #repetition, #frame, #neuron).
+Similar video spike trains should be saved for validating/ testing.
+Then `DataLoader` in `src.utils.py` handles the preprocessing and the data loading.
 
+If you are using a different dataset with a different video resolution the output of the CNN is likely to be different.
+This should be specified using the parameter the `input_latent_size` which is the size of the flattened output of CNN.
 
 #### Forcasting