MTG-pytorch

KO | EN

gitlite — search

#hacktoberfest #ai #rust #deep-learning #linux #docker #nodejs #macos #java #windows #awesome #kubernetes #ios #openai

(★ 12)

Gender/Age attribute grounding using weak supervised manner.

MTG-pytorch 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Visual Cues Grounding Through Weak Suppervision

PyTorch implementation of **[Modularized Textual Grounding for Counterfactual Resilience
](http://openaccess.thecvf.com/content_CVPR_2019/papers/Fang_Modularized_Textual_Grounding_for_Counterfactual_Resilience_CVPR_2019_paper.pdf)** , CVPR 2019.

Qualitative grounding results can be found in our **[webpage](http://www.public.asu.edu/~zfang29/textual_grounding_cvpr2019/website.html)**.

## Introduction
We propose a cross-modal grounding method through weak supervision.

![architecture](https://raw.githubusercontent.com/jacobswan1/MTG-pytorch/master/results/architecture.png "Ground Visual Cue Through a Top-down Guided Design.")

A demonstration on how to load and ground the attribute can be found at : Demo.ipynb

Image --> 'Boy' Attribute -- > 'Lady' Attribute

 <img src="demo_attr.png" width="400" />

## Requirements
1. PyTorch 0.4.
2. Python 3.6.
3. FFT package.

## Dataset
Weakly trained on both COCO or Flickr 30k.

## Usage
Training script for attribute grounding: <pre>Train_attr_attention_embedding.py</pre>

Attention model for attribute grounding, it's based on a pre-trained Res-50 Network on person gender/age classification network:
<pre> /Models/Model7.py</pre>

\<lib> Contains all the neccesary dependencies for our framework, it consists of:

<ul>
 <li>bilinear pooling module: Implemented from <a href="https://github.com/DeepInsight-PCALab/CompactBilinearPooling-Pytorch">Compact Bilinear Pooling</a>. Faster Fourier Transform module is needed before using. Download and install it from <a href="https://github.com/DeepInsight-PCALab/CompactBilinearPooling-Pytorch">here</a> by running:
 <pre>pip3 install pytorch_fft</pre>
</li>
 <li>resnet: We modified the last fully connected layer from 2048d to 256d to a more compact representation.</li>
 <li>nms/roi_align module: Not neccesary in this time. (For entity grounding and bbox detection.)</li>
</ul></pre>

In order to re_train our framework, several things might be modified:
<pre> parser.py </pre>

In parser.py, img_path/annotations need to be changed to your local coco_2017_train directory:
<pre> /path/to/your/local/coco17/image path/annotations/ </pre>

Argument resume is for loading pre-trained overall model.

## Download
To download the pre-trained unsupervised network:
 <ul>
 <li><a href="https://drive.google.com/open?id=10syFqPtkUp4frDV6YEQbgbKs9dUdfTB_">Res50</a> can be found it here.</li>
 <li><a href="https://drive.google.com/open?id=1YPkw0n-beGZ1HTCxxroQTMa21nvg613p"> Model 7 in here</a>.</li>
 </ul>

MTG-pytorch

MTG-pytorch 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

MTG-pytorch

MTG-pytorch 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README