Explainability for Vision Transformers (in PyTorch)

Overview

Explainability for Vision Transformers (in PyTorch)

This repository implements methods for explainability in Vision Transformers.

See also https://jacobgil.github.io/deeplearning/vision-transformer-explainability

Currently implemented:

  • Attention Rollout.

  • Gradient Attention Rollout for class specific explainability. This is our attempt to further build upon and improve Attention Rollout.

  • TBD Attention flow is work in progress.

Includes some tweaks and tricks to get it working:

  • Different Attention Head fusion methods,
  • Removing the lowest attentions.

Usage

  • From code
from vit_grad_rollout import VITAttentionGradRollout

model = torch.hub.load('facebookresearch/deit:main', 
'deit_tiny_patch16_224', pretrained=True)
grad_rollout = VITAttentionGradRollout(model, discard_ratio=0.9, head_fusion='max')
mask = grad_rollout(input_tensor, category_index=243)
  • From the command line:
python vit_explain.py --image_path  --head_fusion  --discard_ratio  --category_index 

If category_index isn't specified, Attention Rollout will be used, otherwise Gradient Attention Rollout will be used.

Notice that by default, this uses the 'Tiny' model from Training data-efficient image transformers & distillation through attention hosted on torch hub.

Where did the Transformer pay attention to in this image?

Image Vanilla Attention Rollout With discard_ratio+max fusion

Gradient Attention Rollout for class specific explainability

The Attention that flows in the transformer passes along information belonging to different classes. Gradient roll out lets us see what locations the network paid attention too, but it tells us nothing about if it ended up using those locations for the final classification.

We can multiply the attention with the gradient of the target class output, and take the average among the attention heads (while masking out negative attentions) to keep only attention that contributes to the target category (or categories).

Where does the Transformer see a Dog (category 243), and a Cat (category 282)?

Where does the Transformer see a Musket dog (category 161) and a Parrot (category 87):

Tricks and Tweaks to get this working

Filtering the lowest attentions in every layer

--discard_ratio

Removes noise by keeping the strongest attentions.

Results for dIfferent values:

Different Attention Head Fusions

The Attention Rollout method suggests taking the average attention accross the attention heads,

but emperically it looks like taking the Minimum value, Or the Maximum value combined with --discard_ratio, works better.

--head_fusion

Image Mean Fusion Min Fusion

References

Requirements

pip install timm

Owner
Jacob Gildenblat
Machine learning / Computer Vision developer.
Jacob Gildenblat
Liecasadi - liecasadi implements Lie groups operation written in CasADi

liecasadi liecasadi implements Lie groups operation written in CasADi, mainly di

Artificial and Mechanical Intelligence 14 Nov 05, 2022
ncnn is a high-performance neural network inference framework optimized for the mobile platform

ncnn ncnn is a high-performance neural network inference computing framework optimized for mobile platforms. ncnn is deeply considerate about deployme

Tencent 16.2k Jan 05, 2023
Generate Cartoon Images using Generative Adversarial Network

AvatarGAN ✨ Generate Cartoon Images using DC-GAN Deep Convolutional GAN is a generative adversarial network architecture. It uses a couple of guidelin

Aakash Jhawar 50 Dec 29, 2022
Nb workflows - A workflow platform which allows you to run parameterized notebooks programmatically

NB Workflows Description If SQL is a lingua franca for querying data, Jupyter sh

Xavier Petit 6 Aug 18, 2022
PyTorch implementation of the wavelet analysis from Torrence & Compo

Continuous Wavelet Transforms in PyTorch This is a PyTorch implementation for the wavelet analysis outlined in Torrence and Compo (BAMS, 1998). The co

Tom Runia 262 Dec 21, 2022
Indoor Panorama Planar 3D Reconstruction via Divide and Conquer

HV-plane reconstruction from a single 360 image Code for our paper in CVPR 2021: Indoor Panorama Planar 3D Reconstruction via Divide and Conquer (pape

sunset 36 Jan 03, 2023
PipeTransformer: Automated Elastic Pipelining for Distributed Training of Large-scale Models

PipeTransformer: Automated Elastic Pipelining for Distributed Training of Large-scale Models This repository is the official implementation of the fol

DistributedML 41 Dec 06, 2022
Code for "3D Human Pose and Shape Regression with Pyramidal Mesh Alignment Feedback Loop"

PyMAF This repository contains the code for the following paper: 3D Human Pose and Shape Regression with Pyramidal Mesh Alignment Feedback Loop Hongwe

Hongwen Zhang 450 Dec 28, 2022
A tool to visualise the results of AlphaFold2 and inspect the quality of structural predictions

AlphaFold Analyser This program produces high quality visualisations of predicted structures produced by AlphaFold. These visualisations allow the use

Oliver Powell 3 Nov 13, 2022
Finite difference solution of 2D Poisson equation. Can handle Dirichlet, Neumann and mixed boundary conditions.

Poisson-solver-2D Finite difference solution of 2D Poisson equation Current version can handle Dirichlet, Neumann, and mixed (combination of Dirichlet

Mohammad Asif Zaman 34 Dec 23, 2022
MNE: Magnetoencephalography (MEG) and Electroencephalography (EEG) in Python

MNE-Python MNE-Python software is an open-source Python package for exploring, visualizing, and analyzing human neurophysiological data such as MEG, E

MNE tools for MEG and EEG data analysis 2.1k Dec 28, 2022
An implementation of an abstract algebra for music tones (pitches).

nbdev template Use this template to more easily create your nbdev project. If you are using an older version of this template, and want to upgrade to

Open Music Kit 0 Oct 10, 2022
On Evaluation Metrics for Graph Generative Models

On Evaluation Metrics for Graph Generative Models Authors: Rylee Thompson, Boris Knyazev, Elahe Ghalebi, Jungtaek Kim, Graham Taylor This is the offic

13 Jan 07, 2023
Code for ECCV 2020 paper "Contacts and Human Dynamics from Monocular Video".

Contact and Human Dynamics from Monocular Video This is the official implementation for the ECCV 2020 spotlight paper by Davis Rempe, Leonidas J. Guib

Davis Rempe 207 Jan 05, 2023
Lorien: A Unified Infrastructure for Efficient Deep Learning Workloads Delivery

Lorien: A Unified Infrastructure for Efficient Deep Learning Workloads Delivery Lorien is an infrastructure to massively explore/benchmark the best sc

Amazon Web Services - Labs 45 Dec 12, 2022
Shared Attention for Multi-label Zero-shot Learning

Shared Attention for Multi-label Zero-shot Learning Overview This repository contains the implementation of Shared Attention for Multi-label Zero-shot

dathuynh 26 Dec 14, 2022
🇰🇷 Text to Image in Korean

KoDALLE Utilizing pretrained language model’s token embedding layer and position embedding layer as DALLE’s text encoder. Background Training DALLE mo

HappyFace 74 Sep 22, 2022
Supporting code for the Neograd algorithm

Neograd This repo supports the paper Neograd: Gradient Descent with a Near-Ideal Learning Rate, which introduces the algorithm "Neograd". The paper an

Michael Zimmer 12 May 01, 2022
Supervised multi-SNE (S-multi-SNE): Multi-view visualisation and classification

S-multi-SNE Supervised multi-SNE (S-multi-SNE): Multi-view visualisation and classification A repository containing the code to reproduce the findings

Theodoulos Rodosthenous 3 Apr 15, 2022
Notebooks, slides and dataset of the CorrelAid Machine Learning Winter School

CorrelAid Machine Learning Winter School Welcome to the CorrelAid ML Winter School! Task The problem we want to solve is to classify trees in Roosevel

CorrelAid 12 Nov 23, 2022