Development kit for MIT Scene Parsing Benchmark

Overview

Development Kit for MIT Scene Parsing Benchmark

[NEW!] Our PyTorch implementation is released in the following repository:

https://github.com/hangzhaomit/semantic-segmentation-pytorch

Introduction

Table of contents:

  • Overview of scene parsing benchmark
  • Benchmark details
    1. Image list and annotations
    2. Submission format
    3. Evaluation routines
  • Pretrained models

Please open an issue for questions, comments, and bug reports.

Overview of Scene Parsing Benchmark

The goal of this benchmark is to segment and parse an image into different image regions associated with semantic categories, such as sky, road, person, and bed. It is similar to semantic segmentation tasks in COCO and Pascal Dataset, but the data is more scene-centric and with a diverse range of object categories. The data for this benchmark comes from ADE20K Dataset (the full dataset will be released after the benchmark) which contains more than 20K scene-centric images exhaustively annotated with objects and object parts. Specifically, the benchmark data is divided into 20K images for training, 2K images for validation, and another batch of held-out images for testing. There are in total 150 semantic categories included in the benchmark for evaluation, which include stuffs like sky, road, grass, and discrete objects like person, car, bed. Note that non-uniform distribution of objects occurs in the images, mimicking a more natural object occurrence in daily scenes.

The webpage of the benchmark is at http://sceneparsing.csail.mit.edu. You could download the data at the webpage.

Benchmark details

Data

There are three types of data, the training, the validation and the testing. The training data contains 20210 images, the validation data contains 2000 images. The testing data contains 2000 images which will be released in middle August. Each image in the training data and validation data has an annotation mask, indicating the labels for each pixel in the image.

After untarring the data file (please download it from http://sceneparsing.csail.mit.edu), the directory structure should be similar to the following,

the training images:

images/training/ADE_train_00000001.jpg
images/training/ADE_train_00000002.jpg
    ...
images/training/ADE_train_00020210.jpg

the corresponding annotation masks for the training images:

annotations/training/ADE_train_00000001.png
annotations/training/ADE_train_00000002.png
    ...
annotations/training/ADE_train_00020210.png

the validation images:

images/validation/ADE_val_00000001.jpg
images/validation/ADE_val_00000002.jpg
    ...
images/validation/ADE_val_00002000.jpg

the corresponding annotation masks for the validation images:

annotations/validation/ADE_val_00000001.png
annotations/validation/ADE_val_00000002.png
    ...
annotations/validation/ADE_val_00002000.png

the testing images will be released in a separate file in the middle Auguest. The directory structure will be: images/testing/ADE_test_00000001.jpg ...

Note: annotations masks contain labels ranging from 0 to 150, where 0 refers to "other objects". We do not consider those pixels in our evaluation.

objectInfo150.txt contains the information about the labels of the 150 semantic categories, including indices, pixel ratios and names.

Submission format to the evaluation server

To evaluate the algorithm on the test set of the benchmark (link: http://sceneparsing.csail.mit.edu/eval/), participants are required to upload a zip file which contains the predicted annotation mask for the given testing images to the evaluation server. The naming of the predicted annotation mask should be the same as the name of the testing images, while the filename extension should be png instead of jpg. For example, the predicted annotation mask for file ADE_test_00000001.jpg should be ADE_test_00000001.png.

Participants should check the zip file to make sure it could be decompressed correctly.

Interclass similarity

Some of the semantic classes in this dataset show some level of visual and semantic similarities across them. In order to quantify such similarities we include a matrix in human_semantic_similarity.mat, which includes human-perceived similarities between the 150 categories and can be used to train the segmentation models. In demoSimilarity.m, we show how to use that file.

Evaluation routines

The performance of the segmentation algorithms will be evaluated by the mean of (1) pixel-wise accuracy over all the labeled pixels, and (2) IoU (intersection over union) avereaged over all the 150 semantic categories.

Intersection over Union = (true positives) / (true positives + false positives + false negatives)
Pixel-wise Accuracy = correctly classifield pixels / labeled pixels
Final score = (Pixel-wise Accuracy + mean(Intersection over Union)) / 2

Demo code

In demoEvaluation.m, we have included our implementation of the standard evaluation metrics (pixel-wise accuracy and IoU) for the benchmark. As mentioned before, we ignore pixels labeled with 0's.

Please change the paths at the begining of the code accordingly to evalutate your own results. While running it correctly, you are expected to see output similar to:

Mean IoU over 150 classes: 0.1000
Pixel-wise Accuracy: 100.00%

In this case, we will take (0.1+1.0)/2=0.55 as your final score.

We have also provided demoVisualization.m, which helps you to visualize individual image results.

Training code

We provide the training code for three popular frameworks, Caffe, Torch7 and PyTorch (https://github.com/CSAILVision/sceneparsing/tree/master/trainingCode). You might need to modify the paths, and the data loader code accordingly to have all the things running on your own computer.

Pre-trained models

We release the pre-trained models for scene parsing at (http://sceneparsing.csail.mit.edu/model/). The demo code along with the model download links is at (https://github.com/CSAILVision/sceneparsing/blob/master/demoSegmentation.m). The models can be used for research only. The detail of how the models are trained is in the reference below. The performance of the models on the validation set of MIT SceneParse150 is as follows,

Prediction

The qualitative results of the models are below:

Prediction

Reference

If you find this scene parse benchmark or the data or the pre-trained models useful, please cite the following paper:

Scene Parsing through ADE20K Dataset. B. Zhou, H. Zhao, X. Puig, S. Fidler, A. Barriuso and A. Torralba. Computer Vision and Pattern Recognition (CVPR), 2017. (http://people.csail.mit.edu/bzhou/publication/scene-parse-camera-ready.pdf)

@inproceedings{zhou2017scene,
    title={Scene Parsing through ADE20K Dataset},
    author={Zhou, Bolei and Zhao, Hang and Puig, Xavier and Fidler, Sanja and Barriuso, Adela and Torralba, Antonio},
    booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
    year={2017}
}

Semantic Understanding of Scenes through ADE20K Dataset. B. Zhou, H. Zhao, X. Puig, S. Fidler, A. Barriuso and A. Torralba. arXiv:1608.05442. (https://arxiv.org/pdf/1608.05442.pdf)

@article{zhou2016semantic,
  title={Semantic understanding of scenes through the ade20k dataset},
  author={Zhou, Bolei and Zhao, Hang and Puig, Xavier and Fidler, Sanja and Barriuso, Adela and Torralba, Antonio},
  journal={arXiv preprint arXiv:1608.05442},
  year={2016}
}
Redash reset for python

redash-reset This will use a default REDASH_SECRET_KEY key of c292a0a3aa32397cdb050e233733900f this allows you to reset the password of the user ID bu

Robert Wiggins 5 Nov 14, 2022
Implementation for "Domain-Specific Bias Filtering for Single Labeled Domain Generalization"

DSBF Introduction This repository contains the implementation code for paper: Domain-Specific Bias Filtering for Single Labeled Domain Generalization

ScottYuan 7 Jan 05, 2023
StyleGAN2-ADA - Official PyTorch implementation

Need Help? If you’re new to StyleGAN2-ADA and looking to get started, please check out this video series from a course Lia Coleman and I taught in Oct

Derrick Schultz 217 Jan 04, 2023
Preparation material for Dropbox interviews

Dropbox-Onsite-Interviews A guide for the Dropbox onsite interview! The Dropbox interview question bank is very small. The bank has been in a Chinese

386 Dec 31, 2022
An open framework for Federated Learning.

Welcome to Intel® Open Federated Learning Federated learning is a distributed machine learning approach that enables organizations to collaborate on m

Intel Corporation 397 Dec 27, 2022
Code release for ConvNeXt model

A ConvNet for the 2020s Official PyTorch implementation of ConvNeXt, from the following paper: A ConvNet for the 2020s. arXiv 2022. Zhuang Liu, Hanzi

Meta Research 4.6k Jan 08, 2023
Official implementation of SIGIR'2021 paper: "Sequential Recommendation with Graph Neural Networks".

SURGE: Sequential Recommendation with Graph Neural Networks This is our TensorFlow implementation for the paper: Sequential Recommendation with Graph

FIB LAB, Tsinghua University 53 Dec 26, 2022
以孤立语假设和宽度优先搜索为基础,构建了一种多通道堆叠注意力Transformer结构的斗地主ai

ddz-ai 介绍 斗地主是一种扑克游戏。游戏最少由3个玩家进行,用一副54张牌(连鬼牌),其中一方为地主,其余两家为另一方,双方对战,先出完牌的一方获胜。 ddz-ai以孤立语假设和宽度优先搜索为基础,构建了一种多通道堆叠注意力Transformer结构的系统,使其经过大量训练后,能在实际游戏中获

freefuiiismyname 88 May 15, 2022
Cross-modal Retrieval using Transformer Encoder Reasoning Networks (TERN). With use of Metric Learning and FAISS for fast similarity search on GPU

Cross-modal Retrieval using Transformer Encoder Reasoning Networks This project reimplements the idea from "Transformer Reasoning Network for Image-Te

Minh-Khoi Pham 5 Nov 05, 2022
Pytorch implementation of "Attention-Based Recurrent Neural Network Models for Joint Intent Detection and Slot Filling"

RNN-for-Joint-NLU Pytorch implementation of "Attention-Based Recurrent Neural Network Models for Joint Intent Detection and Slot Filling"

Kim SungDong 194 Dec 28, 2022
Densely Connected Convolutional Networks, In CVPR 2017 (Best Paper Award).

Densely Connected Convolutional Networks (DenseNets) This repository contains the code for DenseNet introduced in the following paper Densely Connecte

Zhuang Liu 4.5k Jan 03, 2023
Theory-inspired Parameter Control Benchmarks for Dynamic Algorithm Configuration

This repo is for the paper: Theory-inspired Parameter Control Benchmarks for Dynamic Algorithm Configuration The DAC environment is based on the Dynam

Carola Doerr 1 Aug 19, 2022
Python code for the paper How to scale hyperparameters for quickshift image segmentation

How to scale hyperparameters for quickshift image segmentation Python code for the paper How to scale hyperparameters for quickshift image segmentatio

0 Jan 25, 2022
Apollo optimizer in tensorflow

Apollo Optimizer in Tensorflow 2.x Notes: Warmup is important with Apollo optimizer, so be sure to pass in a learning rate schedule vs. a constant lea

Evan Walters 1 Nov 09, 2021
Variational autoencoder for anime face reconstruction

VAE animeface Variational autoencoder for anime face reconstruction Introduction This repository is an exploratory example to train a variational auto

Minzhe Zhang 2 Dec 11, 2021
Deep Image Matting implementation in PyTorch

Deep Image Matting Deep Image Matting paper implementation in PyTorch. Differences "fc6" is dropped. Indices pooling. "fc6" is clumpy, over 100 millio

Yang Liu 724 Dec 27, 2022
HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis

HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis Jungil Kong, Jaehyeon Kim, Jaekyoung Bae In our paper, we p

Rishikesh (ऋषिकेश) 31 Dec 08, 2022
Unofficial implementation of Pix2SEQ

Unofficial-Pix2seq: A Language Modeling Framework for Object Detection Unofficial implementation of Pix2SEQ. Please use this code with causion. Many i

159 Dec 12, 2022
Official PyTorch implementation of RIO

Image-Level or Object-Level? A Tale of Two Resampling Strategies for Long-Tailed Detection Figure 1: Our proposed Resampling at image-level and obect-

NVIDIA Research Projects 17 May 20, 2022
Implementation of paper: "Image Super-Resolution Using Dense Skip Connections" in PyTorch

SRDenseNet-pytorch Implementation of paper: "Image Super-Resolution Using Dense Skip Connections" in PyTorch (http://openaccess.thecvf.com/content_ICC

wxy 114 Nov 26, 2022