Multi-modal Vision Transformers Excel at Class-agnostic Object Detection

Overview

MViTs Excel at Class-agnostic Object Detection

PWC PWC PWC PWC PWC PWC PWC PWC PWC PWC PWC

Multi-modal Vision Transformers Excel at Class-agnostic Object Detection

Muhammad Maaz, Hanoona Rasheed, Salman Khan, Fahad Shahbaz Khan, Rao Muhammad Anwer and Ming-Hsuan Yang

Paper: https://arxiv.org/abs/2111.11430


main figure

Abstract: What constitutes an object? This has been a long-standing question in computer vision. Towards this goal, numerous learning-free and learning-based approaches have been developed to score objectness. However, they generally do not scale well across new domains and for unseen objects. In this paper, we advocate that existing methods lack a top-down supervision signal governed by human-understandable semantics. To bridge this gap, we explore recent Multi-modal Vision Transformers (MViT) that have been trained with aligned image-text pairs. Our extensive experiments across various domains and novel objects show the state-of-the-art performance of MViTs to localize generic objects in images. Based on these findings, we develop an efficient and flexible MViT architecture using multi-scale feature processing and deformable self-attention that can adaptively generate proposals given a specific language query. We show the significance of MViT proposals in a diverse range of applications including open-world object detection, salient and camouflage object detection, supervised and self-supervised detection tasks. Further, MViTs offer enhanced interactability with intelligible text queries.


Architecture overview of MViTs used in this work

Architecture overview


Results


Class-agnostic OD performance of MViTs in comparison with uni-modal detector (RetinaNet) on several datasets. MViTs show consistently good results on all datasets.

Results


Enhanced Interactability: Effect of using different intuitive text queries on the MDef-DETR class-agnostic OD performance. Combining detections from multiple queries captures varying aspects of objectness.

Results


Generalization to Rare/Novel Classes: MDef-DETR class-agnostic OD performance on rarely and frequently occurring categories in the pretraining captions. The numbers on top of the bars indicate occurrences of the corresponding category in the training dataset. The MViT achieves good recall values even for the classes with no or very few occurrences.

Results


Open-world Object Detection: Effect of using class-agnostic OD proposals from MDef-DETR for pseudo labelling of unknowns in Open World Detector (ORE).

Results


Pretraining for Class-aware Object Detection: Effect of using MDef-DETR proposals for pre-training of DETReg instead of Selective Search proposals.

Results


Evaluation

The provided codebase contains the pre-computed detections for all datasets using ours MDef-DETR model. The provided directory structure is as follows,

-> README.md
-> LICENSE
-> get_eval_metrics.py
-> get_multi_dataset_eval_metrics.py
-> data
    -> voc2007
        -> combined.pkl
    -> coco
        -> combined.pkl
    -> kitti
        -> combined.pkl
    -> kitchen
        -> combined.pkl
    -> cliaprt
        -> combined.pkl
    -> comic
        -> combined.pkl
    -> watercolor
        -> combined.pkl
    -> dota
        -> combined.pkl

Where combined.pkl contains the combined detections from multiple intutive text queries for corresponding datasets. (Refer Section 5.1: Enhanced Interactability for more details)

Download the annotations for all datasets and arrange them as shown below. Note that the script expect COCO annotations in standard COCO format & annotations of all other datasets in VOC format.

...
...
-> data
    -> voc2007
        -> combined.pkl
        -> Annotations
    -> coco
        -> combined.pkl
        -> instances_val2017_filt.json
    -> kitti
        -> combined.pkl
        -> Annotations
        ...
    -> kitchen
        -> combined.pkl
        -> Annotations
    -> cliaprt
        -> combined.pkl
        -> Annotations
    -> comic
        -> combined.pkl
        -> Annotations
    -> watercolor
        -> combined.pkl
        -> Annotations
    -> dota
        -> combined.pkl
        -> Annotations

Once the above mentioned directory structure is created, follow the following steps to calculate the metrics.

  1. Install numpy
$ pip install numpy
  1. Calculate metrics
$ python get_multi_dataset_eval_metrics.py

The calculated metrics will be stored in a data.csv file in the same directory.


Citation

If you use our work, please consider citing:

@article{Maaz2021Multimodal,
    title={Multi-modal Transformers Excel at Class-agnostic Object Detection},
    author={Muhammad Maaz and Hanoona Rasheed and Salman Khan and Fahad Shahbaz Khan and Rao Muhammad Anwer and Ming-Hsuan Yang},
    journal={ArXiv 2111.11430},
    year={2021}
}

Contact

Should you have any question, please contact [email protected] or [email protected]

🚀 Note: The repository contains the minimum evaluation code. The complete training and inference scripts along with pretrained models will be released soon. Stay Tuned!

Comments
  • aligning image text pairs

    aligning image text pairs

    I have a question on the paper: you train on aligned image-text pairs. How do you create this alignment? is it the same way as in MDeTr? I did not fully understand from the paper, especially for non-natural images like satellite images or medical images.

    opened by nikky4D 6
  • Loading checkpoints for inference

    Loading checkpoints for inference

    Which checkpoints in drive link you provided will load correctly in default MDefDETR model without any errors? Im getting missing/unexpected keys errors.

    documentation 
    opened by KaleemW 4
  • Is EMA used in this work?

    Is EMA used in this work?

    Hello author, thanks for your great work. I raise a question about the usage of Exponential Moving Average (EMA) in this paper, hoping you can provide me with some clues. It seems that this paper does not detail in this part. As far as I know, MDETR uses it and evaluate use the EMA model. So I wonder is it used in this work? If it is actually used, why should we evaluate by the EMA model rather than the original one?

    opened by JacobYuan7 4
  • one of the variables needed for gradient computation has been modified by an inplace operation

    one of the variables needed for gradient computation has been modified by an inplace operation

    RuntimeError: one of the variables needed for gradient computation has been modified by an inplace operation: [torch.cuda.LongTensor [2, 20]] is at version 3; expected version 2 instead. Hint: enable anomaly detection to find the operation that failed to compute its gradient, with torch.autograd.set_detect_anomaly(True).

    This error will terminate the training procedure when training mdef_detr using the PyTorch environment as you advise(torch==1.8.0+cu111).

    And I found the variables of 'transformer.text_encoder.pooler.dense.weight' does not have grad. This may be the main reason for this error.

    opened by xushilin1 2
  • Loading the Faster RCNN checkpoint

    Loading the Faster RCNN checkpoint

    Greetings

    The readme states: (Feb 01, 2022) Training codes for MDef-DETR and MDef-DETR minus Language models are released -> training/README.md Instructions to use class-agnostic object detection behavior of MDef-DETR on different applications are released -> applications/README.md All the pretrained models (MDef-DETR, Def-DETR, MDETR, DETReg, Faster-RCNN, RetinaNet, ORE, and others), along with the instructions to reproduce the results are released -> this link

    Following the link to the google drive, only provides me with the model weight for the Faster-RCNN, but not with instructions on how to load it and which framework to use. I have tried creating a Faster-RCNN-resnet101 model with pytorch, but when I load the model weight, it states that the layer names does not match. Any guidance would be much appreciated.

    Best regards Martin

    uni-modal-detectors 
    opened by MartinPedersenpp 2
  • Need to understand how to import weights

    Need to understand how to import weights

    Hello,

    Firstly, I'd like to congratulate you for bringing this amazing work. Class agnostic object detection is much needed currently in the industry and this would be a great way to solve the problem.

    I wanted to test your model on some custom data. However, I cannot import pre-trained weights from the link you have provided. I can see the zip file but I couldn't find a way to import them. I'm using OpenCV to import weights. It is asking me to have a config file as well as .weights file.

    Could you please help me which library to use to import weights when I'm working on a jupyter notebook?

    Thank you,

    opened by abhi-vellala 2
  • pretrain data download

    pretrain data download

    if is it possible to split pretrain data into multiple seperate zip files。 I download data from google drive : https://drive.google.com/drive/folders/1-3kAsyZIVFbNelRXrF93Y5tMgOypv2jV i cannot download this data because of google drive time limit(less than 1 hours) and my limit network bandwidth。

    documentation 
    opened by zhouxingguang 1
  • Training code release

    Training code release

    This pull request adds

    • Training codes for MDef-DETR and MDef-DETR minus Language models
    • Instructions to use class-agnostic object detection behavior of MDef-DETR on different applications
    • All the pre-trained models (MDef-DETR, Def-DETR, MDETR, DETReg, Faster-RCNN, RetinaNet, ORE, and others), along with the instructions to reproduce the results
    opened by mmaaz60 0
  • Questions about your training procedure?

    Questions about your training procedure?

    To my understanding, I think you use image-text pairs as inputs and only bbox annotations as supervision signals without any class labels, does it right?

    opened by GYslchen 1
  • Questions about your pretrained model

    Questions about your pretrained model

    Does the pre-trained model you provide cover the categories on LVIS data? If I want to do open-world object detection on the LVIS dataset, can I directly use your pre-trained model to generate the proposals or should I need to filter the dataset so that it doesn't contain any object in the LVIS dataset?

    opened by chengsilin 1
  • how to generate 'tokens_positive'  ann from detector dataset like object365?

    how to generate 'tokens_positive' ann from detector dataset like object365?

    I found 'tokens_positive' was used in your ann file. could you please release the code of how to process detect data like coco to get the 'tokens_positive' ann results?

    documentation 
    opened by zhouxingguang 1
Releases(v1.0)
  • v1.0(Feb 1, 2022)

    • Training codes for MDef-DETR and MDef-DETR minus Language models are released -> training/README.md
    • Instructions to use class-agnostic object detection behavior of MDef-DETR on different applications are released -> applications/README.md
    • All the pretrained models (MDef-DETR, Def-DETR, MDETR, DETReg, Faster-RCNN, RetinaNet, ORE, and others), along with the instructions to reproduce the results are released -> this link
    Source code(tar.gz)
    Source code(zip)
  • v0.1(Nov 25, 2021)

    Evaluation Code & Pre-trained Models

    • Releases evaluation code for MDef-DETR and 'MDef-DETR w/o Language Branch' model
    • Releases the pre-trained weights for both models
    • Releases the pre-computed predictions for both the models
    Source code(tar.gz)
    Source code(zip)
Owner
Muhammad Maaz
An Electrical Engineer with experience in Computer Vision software development. Skilled in Machine Learning, Deep Learning and Computer Vision.
Muhammad Maaz
Simple image captioning model - CLIP prefix captioning.

Simple image captioning model - CLIP prefix captioning.

688 Jan 04, 2023
DETReg: Unsupervised Pretraining with Region Priors for Object Detection

DETReg: Unsupervised Pretraining with Region Priors for Object Detection Amir Bar, Xin Wang, Vadim Kantorov, Colorado J Reed, Roei Herzig, Gal Chechik

Amir Bar 283 Dec 27, 2022
A short and easy PyTorch implementation of E(n) Equivariant Graph Neural Networks

Simple implementation of Equivariant GNN A short implementation of E(n) Equivariant Graph Neural Networks for HOMO energy prediction. Just 50 lines of

Arsenii Senya Ashukha 97 Dec 23, 2022
[CVPR 2022] Unsupervised Image-to-Image Translation with Generative Prior

GP-UNIT - Official PyTorch Implementation This repository provides the official PyTorch implementation for the following paper: Unsupervised Image-to-

Shuai Yang 125 Jan 03, 2023
QHack—the quantum machine learning hackathon

Official repo for QHack—the quantum machine learning hackathon

Xanadu 72 Dec 21, 2022
Advancing mathematics by guiding human intuition with AI

Advancing mathematics by guiding human intuition with AI This repo contains two colab notebooks which accompany the paper, available online at https:/

DeepMind 315 Dec 26, 2022
Ego4d dataset repository. Download the dataset, visualize, extract features & example usage of the dataset

Ego4D EGO4D is the world's largest egocentric (first person) video ML dataset and benchmark suite, with 3,600 hrs (and counting) of densely narrated v

Meta Research 118 Jan 07, 2023
Background Matting: The World is Your Green Screen

Background Matting: The World is Your Green Screen By Soumyadip Sengupta, Vivek Jayaram, Brian Curless, Steve Seitz, and Ira Kemelmacher-Shlizerman Th

Soumyadip Sengupta 4.6k Jan 04, 2023
a practicable framework used in Deep Learning. So far UDL only provide DCFNet implementation for the ICCV paper (Dynamic Cross Feature Fusion for Remote Sensing Pansharpening)

UDL UDL is a practicable framework used in Deep Learning (computer vision). Benchmark codes, results and models are available in UDL, please contact @

Xiao Wu 11 Sep 30, 2022
Demo for the paper "Overlap-aware low-latency online speaker diarization based on end-to-end local segmentation"

Streaming speaker diarization Overlap-aware low-latency online speaker diarization based on end-to-end local segmentation by Juan Manuel Coria, Hervé

Juanma Coria 187 Jan 06, 2023
Understanding the Generalization Benefit of Model Invariance from a Data Perspective

Understanding the Generalization Benefit of Model Invariance from a Data Perspective This is the code for our NeurIPS2021 paper "Understanding the Gen

1 Jan 15, 2022
Privacy-Preserving Portrait Matting [ACM MM-21]

Privacy-Preserving Portrait Matting [ACM MM-21] This is the official repository of the paper Privacy-Preserving Portrait Matting. Jizhizi Li∗, Sihan M

Jizhizi_Li 212 Dec 27, 2022
Convolutional Neural Network for Text Classification in Tensorflow

This code belongs to the "Implementing a CNN for Text Classification in Tensorflow" blog post. It is slightly simplified implementation of Kim's Convo

Denny Britz 5.5k Jan 02, 2023
This is the PyTorch implementation of GANs N’ Roses: Stable, Controllable, Diverse Image to Image Translation

Official PyTorch repo for GAN's N' Roses. Diverse im2im and vid2vid selfie to anime translation.

1.1k Jan 01, 2023
Using deep learning model to detect breast cancer.

Breast-Cancer-Detection Breast cancer is the most frequent cancer among women, with around one in every 19 women at risk. The number of cases of breas

1 Feb 13, 2022
This repository contains several image-to-image translation models, whcih were tested for RGB to NIR image generation. The models are Pix2Pix, Pix2PixHD, CycleGAN and PointWise.

RGB2NIR_Experimental This repository contains several image-to-image translation models, whcih were tested for RGB to NIR image generation. The models

5 Jan 04, 2023
This is an open source library implementing hyperbox-based machine learning algorithms

hyperbox-brain is a Python open source toolbox implementing hyperbox-based machine learning algorithms built on top of scikit-learn and is distributed

Complex Adaptive Systems (CAS) Lab - University of Technology Sydney 21 Dec 14, 2022
A Deep learning based streamlit web app which can tell with which bollywood celebrity your face resembles.

Project Name: Which Bollywood Celebrity You look like A Deep learning based streamlit web app which can tell with which bollywood celebrity your face

BAPPY AHMED 20 Dec 28, 2021
Real-time pose estimation accelerated with NVIDIA TensorRT

trt_pose Want to detect hand poses? Check out the new trt_pose_hand project for real-time hand pose and gesture recognition! trt_pose is aimed at enab

NVIDIA AI IOT 803 Jan 06, 2023
Simple Tensorflow implementation of Toward Spatially Unbiased Generative Models (ICCV 2021)

Spatial unbiased GANs — Simple TensorFlow Implementation [Paper] : Toward Spatially Unbiased Generative Models (ICCV 2021) Abstract Recent image gener

Junho Kim 16 Apr 15, 2022