update requirement; provide the link to download v1-dataset

README.md

@@ -10,6 +10,8 @@ Accepted for publication at the NeurIPS conference 2021
 
 ## Environment
 
+`conda create -n rsnn_SM python=3.8`
+`conda activate rsnn_SM`
 `pip install -r requirements.txt`
 
 ## Execution example
@@ -29,12 +31,14 @@ Overall, the network `src.buildnet.py` consists of two parts. One is the CNN `co
 
 `DataLoader` in the `src.utils.py` handles preprocessing and loading data.
 
-To run the code, one will need to prepare:
+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 may want to tailor the CNN that pre-processes the video according to the input.
+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 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.
 
 

requirements.txt

@@ -1,8 +1,8 @@
 matplotlib==3.4.3
-numpy==1.19.5
+numpy==1.18.5
 six==1.15.0
 scikit-image==0.18.3
 pandas==1.1.5
 seaborn==0.11.2
 scipy==1.7.1
-tensorflow==2.6.0
+tensorflow==2.3.1