cleanup repo

dunnkers · Oct 29, 2020 · 0a93db6 · 0a93db6
1 parent bf786ab
commit 0a93db6
Show file tree

Hide file tree

Showing 38 changed files with 1,143 additions and 25,016 deletions.
diff --git a/.gitignore b/.gitignore
@@ -1 +1,2 @@
-.DS_Store
+.DS_Store
+/latent-backdoor/data
diff --git a/latent-backdoor/README.md b/latent-backdoor/README.md
@@ -7,10 +7,19 @@ Make sure to have Docker daemon running.
 
 Before you start development: download the test and validation datasets at http://www.image-net.org/download-images (ISVRC 2012 dataset) or follow `https://github.com/onnx/models/blob/master/vision/classification/imagenet_prep.md` 
 
+## Training locally
+Download the datasets for teacher, poisoned teacher and student model:
+
+ - Teacher: https://drive.google.com/file/d/19tfqPEx8jzRaj5qw_tF4lbQ2aUWMyZqH/view?usp=sharing
+ - Poisoned teacher: https://drive.google.com/file/d/1_pXhJrT1L0ksiV0_hIqUFKKP1x9PP29Z/view?usp=sharing
+ - Student: https://drive.google.com/file/d/1bFI6Np2L6HzMCu3nZa3425wfDgojpKSt/view?usp=sharing
+
+Start the app: `docker-compose up` and then follow the link shown in the console. In the notebooks folder you will find a jupyter notebook called `train_mobilenet.ipynb`. To use a trained model, see the `eval_mobilenet.ipynb` notebook instead. This will also produce a `my-model.onnx` file, which should be stored in `/home/checkpoints`. You can then find the model on your own machine in this repo in the `/latent-backdoor/params` folder.
+
 ## Running on peregrine
 
  1. ensure that the GPU version of mxnet is enabled in `latent-backdoor/train/setup-dependencies.sh`
  2. clone the repository into your home directory on peregrine
- 3. submit the batch: `sbatch train-peregrine.txt`
-
- The batch will ensure that all training data is available in `/data/<s-number>/data`. It will also install all required dependencies on the node its running on
+ 3. run `pip install gdown`
+ 4. Update the files `ensure-data.sh`, `train-teacher.py`, `train-trigger.py` and `train-student.py` to point to your own `/data/<s or p number>` folder. Then run it to download and extract the data into the appropriate folder
+ 5. submit the batch for the network that you want to train: `sbatch train-peregrine.txt`
diff --git a/latent-backdoor/active-users-info.txt b/latent-backdoor/active-users-info.txt
diff --git a/latent-backdoor/data/test/img.jpg b/latent-backdoor/data/test/img.jpg
diff --git a/latent-backdoor/data/test/img.png b/latent-backdoor/data/test/img.png
diff --git a/latent-backdoor/docker-compose.yml b/latent-backdoor/docker-compose.yml
@@ -5,7 +5,6 @@ services:
     ports:
       - "8888:8888"
     volumes:
-      - "./data/tmp:/home/data"
+      - "./data:/home/data"
       - "./notebooks:/scripts/notebooks"
-      - "./train/params:/home/checkpoints"
-
+      - "./params:/home/checkpoints"
diff --git a/latent-backdoor/train/ensure-data.sh → latent-backdoor/ensure-data.sh b/latent-backdoor/train/ensure-data.sh → latent-backdoor/ensure-data.sh
diff --git a/latent-backdoor/notebooks/.ipynb_checkpoints/Untitled-checkpoint.ipynb b/latent-backdoor/notebooks/.ipynb_checkpoints/Untitled-checkpoint.ipynb
@@ -0,0 +1,6 @@
+{
+ "cells": [],
+ "metadata": {},
+ "nbformat": 4,
+ "nbformat_minor": 4
+}
diff --git a/latent-backdoor/notebooks/demo.ipynb → .../.ipynb_checkpoints/demo-checkpoint.ipynb b/latent-backdoor/notebooks/demo.ipynb → .../.ipynb_checkpoints/demo-checkpoint.ipynb
@@ -2,13 +2,17 @@
  "cells": [
   {
    "cell_type": "code",
-   "execution_count": 17,
+   "execution_count": 1,
    "metadata": {},
    "outputs": [
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
+      "Download 'mobilenetv2-1.0.onnx'...\n",
+      "Done.\n",
+      "Download 'imagenet_class_index.json'...\n",
+      "Done.\n",
       "(1, 3, 224, 224)\n",
       "(1, 3, 224, 224) 0.0 1.0\n",
       "(1000,)\n",

diff --git a/latent-backdoor/notebooks/.ipynb_checkpoints/eval_mobilenet-checkpoint.ipynb b/latent-backdoor/notebooks/.ipynb_checkpoints/eval_mobilenet-checkpoint.ipynb