README

IntelLabs · Aug 19, 2021 · c1e01b4 · c1e01b4
1 parent 8b1cbae
commit c1e01b4
Showing 1 changed file with 36 additions and 4 deletions.
diff --git a/CLOC/README.md b/CLOC/README.md
@@ -2,31 +2,63 @@
 
 # Code for ICCV 2021 paper: Online Continual Learning with Natural Distribution Shifts: An Empirical Study with Visual Data. 
 
+Author: Zhipeng Cai, Ozan Sener, Vladlen Koltun
+
 [(Paper link)](coming soon)
 
 This repo contains the code for reproducing the experiments in the paper. 
 
 Prerequisite
 ============
 python 3
+
 pytorch 1.7+
+
 tensorboardX
+
 numpy
 
 Usage
 =====
-1. Data preparation:
+1. clone this repo.
 
+2. download meta-data of CLOC from "Put the link here". Decompress it into "CLOC/data_preparation/release", via:
 
+mv metadata.tar.gz CLOC/data_preparation/
+tar -xvzf metadata.tar.gz
 
-About
-=====
+3. download images (change the root_folder and other parameters in the download_images.py file if you like, an important function for download_images.py is to simultaneously download different parts of the dataset, please see the file for more details), via:
+
+cd CLOC/data_preparation/download_images
+python /download_images.py
+
+4. run the experiments:
+
+simply go to exp_<name_of_experiments> folders, and run the experiment that you want to replicate. "exp_best_model" refers to the code to train the proposed OCL model, i.e., using PoLRS, ADRep and small batch sizes.
+
+In each exp_<name_of_experiments> folder, there will be pairs of scripts named as "train_xxx.sh" and "eval_xxx.sh". Both can be run by simply type in:
+
+bash xxx_xxx.sh
+
+The train_xxx.sh file trains the OCL model and plots the average online accuracy.
+
+The eval_xxx.sh produces the backward transfer curve.
+
+The time axis for different plots may have a scaling effect, one can convert the time by simply normalize the time axis into [0,1] and then mutiply individual time point by the maximum number of images or the maximum wall-clock time. 
+
+5. use tensorboard to monitor the results
+
+go to each exp_xxx folder, do:
+
+tensorboard --logdir=./
+
+The output folder of individual experiments can be found in the first line of the output logs.
 
 ------------------------
 Contact
 ------------------------
 
-Homepage:https://zhipengcai.github.io/
+Homepage: https://zhipengcai.github.io/
 
 Email: czptc2h@gmail.com