Reverse engineer your pytorch vision models, in style

Related tags

Deep Learningrover
Overview

🔍 Rover

Reverse engineer your CNNs, in style

Open In Colab

Rover will help you break down your CNN and visualize the features from within the model. No need to write weirdly abstract code to visualize your model's features anymore.

💻 Usage

git clone https://github.com/Mayukhdeb/rover.git; cd rover

install requirements:

pip install -r requirements.txt
from rover import core
from rover.default_models import models_dict

core.run(models_dict = models_dict)

and then run the script with streamlit as:

$ streamlit run your_script.py

if everything goes right, you'll see something like:

You can now view your Streamlit app in your browser.

  Local URL: http://localhost:8501

🧙 Custom models

rover supports pretty much any PyTorch model with an input of shape [N, 3, H, W] (even segmentation models/VAEs and all that fancy stuff) with imagenet normalization on input.

import torchvision.models as models 
model = models.resnet34(pretrained= True)  ## or any other model (need not be from torchvision.models)

models_dict = {
    'my model': model,  ## add in any number of models :)
}

core.run(
    models_dict = models_dict
)

🖼️ Channel objective

Optimizes a single channel from one of the layer(s) selected.

  • layer index: specifies which layer you want to use out of the layers selected.
  • channel index: specifies the exact channel which needs to be visualized.

🧙‍♂️ Writing your own objective

This is for the smarties who like to write their own objective function. The only constraint is that the function should be named custom_func.

Here's an example:

def custom_func(layer_outputs):
    '''
    layer_outputs is a list containing 
    the outputs (torch.tensor) of each layer you selected

    In this example we'll try to optimize the following:
    * the entire first layer -> layer_outputs[0].mean()
    * 20th channel of the 2nd layer -> layer_outputs[1][20].mean()
    '''
    loss = layer_outputs[0].mean() + layer_outputs[1][20].mean()
    return -loss

Running on google colab

Check out this notebook. I'll also include the instructions here just in case.

Clone the repo + install dependencies

!git clone https://github.com/Mayukhdeb/rover.git
!pip install torch-dreams --quiet
!pip install streamlit --quiet

Navigate into the repo

import os 
os.chdir('rover')

Write your file into a script from a cell. Here I wrote it into test.py

%%writefile  test.py

from rover import core
from rover.default_models import models_dict

core.run(models_dict = models_dict)

Run script on a thread

import threading

proc = threading.Thread(target= os.system, args=['streamlit run test.py'])
proc.start()

Download ngrok:

!wget https://bin.equinox.io/c/4VmDzA7iaHb/ngrok-stable-linux-amd64.zip
!unzip -o ngrok-stable-linux-amd64.zi

More ngrok stuff

get_ipython().system_raw('./ngrok http 8501 &')

Get your URL where rover is hosted

!curl -s http://localhost:4040/api/tunnels | python3 -c \
    "import sys, json; print(json.load(sys.stdin)['tunnels'][0]['public_url'])"

💻 Args

  • width (int, optional): Width of image to be optimized
  • height (int, optional): Height of image to be optimized
  • iters (int, optional): Number of iterations, higher -> stronger visualization
  • lr (float, optional): Learning rate
  • rotate (deg) (int, optional): Max rotation in default transforms
  • scale max (float, optional): Max image size factor.
  • scale min (float, optional): Minimum image size factor.
  • translate (x) (float, optional): Maximum translation factor in x direction
  • translate (y) (float, optional): Maximum translation factor in y direction
  • weight decay (float, optional): Weight decay for default optimizer. Helps prevent high frequency noise.
  • gradient clip (float, optional): Maximum value of the norm of gradient.

Run locally

Clone the repo

git clone https://github.com/Mayukhdeb/rover.git

install requirements

pip install -r requirements.txt

showtime

streamlit run test.py
Owner
Mayukh Deb
Learning about life, one epoch at a time
Mayukh Deb
Contains code for the paper "Vision Transformers are Robust Learners".

Vision Transformers are Robust Learners This repository contains the code for the paper Vision Transformers are Robust Learners by Sayak Paul* and Pin

Sayak Paul 103 Jan 05, 2023
Shape Matching of Real 3D Object Data to Synthetic 3D CADs (3DV project @ ETHZ)

Real2CAD-3DV Shape Matching of Real 3D Object Data to Synthetic 3D CADs (3DV project @ ETHZ) Group Member: Yue Pan, Yuanwen Yue, Bingxin Ke, Yujie He

24 Jun 22, 2022
Simulation of the solar system using various nummerical methods

solar-system Simulation of the solar system using various nummerical methods Download the repo Make shure matplotlib, scipy etc. are installed execute

Caspar 7 Jul 15, 2022
Keras community contributions

keras-contrib : Keras community contributions Keras-contrib is deprecated. Use TensorFlow Addons. The future of Keras-contrib: We're migrating to tens

Keras 1.6k Dec 21, 2022
ADGAN - The Implementation of paper Controllable Person Image Synthesis with Attribute-Decomposed GAN

ADGAN - The Implementation of paper Controllable Person Image Synthesis with Attribute-Decomposed GAN CVPR 2020 (Oral); Pose and Appearance Attributes Transfer;

Men Yifang 400 Dec 29, 2022
💡 Type hints for Numpy

Type hints with dynamic checks for Numpy! (❒) Installation pip install nptyping (❒) Usage (❒) NDArray nptyping.NDArray lets you define the shape and

Ramon Hagenaars 377 Dec 28, 2022
CCCL: Contrastive Cascade Graph Learning.

CCGL: Contrastive Cascade Graph Learning This repo provides a reference implementation of Contrastive Cascade Graph Learning (CCGL) framework as descr

Xovee Xu 19 Dec 05, 2022
This project deals with the detection of skin lesions within the ISICs dataset using YOLOv3 Object Detection with Darknet.

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. Skin Lesion detection using YOLO This project deal

Lalith Veerabhadrappa Badiger 1 Nov 22, 2021
Learning Continuous Signed Distance Functions for Shape Representation

DeepSDF This is an implementation of the CVPR '19 paper "DeepSDF: Learning Continuous Signed Distance Functions for Shape Representation" by Park et a

Meta Research 1.1k Jan 01, 2023
Code to reproduce experiments in the paper "Explainability Requires Interactivity".

Explainability Requires Interactivity This repository contains the code to train all custom models used in the paper Explainability Requires Interacti

Digital Health & Machine Learning 5 Apr 07, 2022
Image-to-image translation with conditional adversarial nets

pix2pix Project | Arxiv | PyTorch Torch implementation for learning a mapping from input images to output images, for example: Image-to-Image Translat

Phillip Isola 9.3k Jan 08, 2023
Reference PyTorch implementation of "End-to-end optimized image compression with competition of prior distributions"

PyTorch reference implementation of "End-to-end optimized image compression with competition of prior distributions" by Benoit Brummer and Christophe

Benoit Brummer 6 Jun 16, 2022
[ACL-IJCNLP 2021] Improving Named Entity Recognition by External Context Retrieving and Cooperative Learning

CLNER The code is for our ACL-IJCNLP 2021 paper: Improving Named Entity Recognition by External Context Retrieving and Cooperative Learning CLNER is a

71 Dec 08, 2022
Feed forward VQGAN-CLIP model, where the goal is to eliminate the need for optimizing the latent space of VQGAN for each input prompt

Feed forward VQGAN-CLIP model, where the goal is to eliminate the need for optimizing the latent space of VQGAN for each input prompt. This is done by

Mehdi Cherti 135 Dec 30, 2022
Code for ICE-BeeM paper - NeurIPS 2020

ICE-BeeM: Identifiable Conditional Energy-Based Deep Models Based on Nonlinear ICA This repository contains code to run and reproduce the experiments

Ilyes Khemakhem 65 Dec 22, 2022
Repository of continual learning papers

Continual learning paper repository This repository contains an incomplete (but dynamically updated) list of papers exploring continual learning in ma

29 Jan 05, 2023
Code for paper: Towards Tokenized Human Dynamics Representation

Video Tokneization Codebase for video tokenization, based on our paper Towards Tokenized Human Dynamics Representation. Prerequisites (tested under Py

Kenneth Li 20 May 31, 2022
Official implementation of ACTION-Net: Multipath Excitation for Action Recognition (CVPR'21).

ACTION-Net Official implementation of ACTION-Net: Multipath Excitation for Action Recognition (CVPR'21). Getting Started EgoGesture data folder struct

V-Sense 171 Dec 26, 2022
BADet: Boundary-Aware 3D Object Detection from Point Clouds (Pattern Recognition 2022)

BADet: Boundary-Aware 3D Object Detection from Point Clouds (Pattern Recognition

Rui Qian 17 Dec 12, 2022
Deep Learning and Logical Reasoning from Data and Knowledge

Logic Tensor Networks (LTN) Logic Tensor Network (LTN) is a neurosymbolic framework that supports querying, learning and reasoning with both rich data

171 Dec 29, 2022