Graph Rcnn.pytorch

Pytorch code for our ECCV 2018 paper "Graph R-CNN for Scene Graph Generation" and other papers
Alternatives To Graph Rcnn.pytorch
Project NameStarsDownloadsRepos Using ThisPackages Using ThisMost Recent CommitTotal ReleasesLatest ReleaseOpen IssuesLicenseLanguage
Autolod1,354
8 months ago7otherC#
Automatic LOD generation + scene optimization
Scene Graph Benchmark.pytorch760
10 months ago100mitJupyter Notebook
A new codebase for popular Scene Graph Generation methods (2020). Visualization & Scene Graph Extraction on custom images/datasets are provided. It's also a PyTorch implementation of paper “Unbiased Scene Graph Generation from Biased Training CVPR 2020”
Three.terrain50954a year ago6July 06, 20213mitJavaScript
A procedural terrain generation engine for use with the Three.js 3D graphics library for the web.
Graph Rcnn.pytorch482
3 years ago32Python
Pytorch code for our ECCV 2018 paper "Graph R-CNN for Scene Graph Generation" and other papers
Factorizablenet157
4 years ago6Python
Factorizable Net (Multi-GPU version): An Efficient Subgraph-based Framework for Scene Graph Generation
Scene_generation145
3 years ago6apache-2.0Python
A PyTorch implementation of the paper: Specifying Object Attributes and Relations in Interactive Scene Generation
Text2scene101
a year ago2Python
[CVPR 2019] Text2Scene: Generating Compositional Scenes from Textual Descriptions
Proceduralpatterns2d57
5 years agootherC#
Scfusion41
10 months agobsd-2-clauseC++
The code of SCFusion: Real-time Incremental Scene Reconstruction with Semantic Completion will be released soon.
Ebm_compositionality31
a year ago3Python
[NeurIPS'20] Code for the Paper Compositional Visual Generation and Inference with Energy Based Models
Alternatives To Graph Rcnn.pytorch
Select To Compare


Alternative Project Comparisons
Readme

graph-rcnn.pytorch

Pytorch code for our ECCV 2018 paper "Graph R-CNN for Scene Graph Generation"

Introduction

This project is a set of reimplemented representative scene graph generation models based on Pytorch 1.0, including:

Our reimplementations are based on the following repositories:

Why we need this repository?

The goal of gathering all these representative methods into a single repo is to establish a more fair comparison across different methods under the same settings. As you may notice in recent literatures, the reported numbers for IMP, MSDN, Graph R-CNN and Neural Motifs are usually confusing, especially due to the big gap between IMP style methods (first three) and Neural Motifs-style methods (neural motifs paper and other variants built on it). We hope this repo can establish a good benchmark for various scene graph generation methods, and contribute to the research community!

Checklist

  • [x] Faster R-CNN Baseline (🎈 2019-07-04)
  • [x] Scene Graph Generation Baseline (🎈 2019-07-06)
  • [x] Iterative Message Passing (IMP) (🎈 2019-07-07)
  • [x] Multi-level Scene Description Network (MSDN:no region caption) (🎈 2019-08-24)
  • [x] Neural Motif (Frequency Prior Baseline) (🎈 2019-07-08)
  • [x] Graph R-CNN (w/o relpn, GCNs) (🎈 2019-08-24)
  • [x] Graph R-CNN (w relpn, GCNs) (🎈 2020-01-13)
  • [ ] Graph R-CNN (w relpn, aGCNs)
  • [ ] Neural Motif
  • [ ] RelDN (Graphical Contrastive Losses)

Benchmarking

Object Detection

source backbone model bs lr lr_decay [email protected] [email protected]:0.95
this repo Res-101 faster r-cnn 6 5e-3 70k,90k 24.8 12.8

Scene Graph Generation (Frequency Prior Only)

source backbone model bs lr lr_decay [email protected] [email protected] [email protected]
this repo Res-101 freq 6 5e-3 70k,90k 19.4 25.0 28.5
motifnet VGG-16 freq - - - 17.7 23.5 27.6

* freq = frequency prior baseline

Scene Graph Generation (Joint training)

source backbone model bs lr lr_decay [email protected] [email protected] [email protected]
this repo Res-101 vanilla 6 5e-3 70k,90k 10.4 14.3 16.8

Scene Graph Generation (Step training)

source backbone model bs lr [email protected] [email protected] [email protected] [email protected]
this repo Res-101 vanilla 8 5e-3 24.2 10.5 13.8 16.1
this repo Res-101 imp 8 5e-3 24.2 16.7 21.7 25.2
motifnet VGG-16 imp - - - 14.6 20.7 24.5

* you can click 'this repo' in above table to download the checkpoints.

The above table shows that our reimplementation of baseline and imp algorithm match the performance reported in mofitnet.

Comparisons with other Methods

model bs lr [email protected] [email protected] [email protected] [email protected]
vanilla 8 5e-3 24.2 10.5 13.8 16.1
imp 8 5e-3 24.2 16.7 21.7 25.2
msdn 8 5e-3 24.2 18.3 23.6 27.1
graph-rcnn(no att) 8 5e-3 24.2 18.8 23.7 26.2

* you can click 'model' in above table to download the checkpoints.

Accordingly, all models achieved significantly better numbers compared with those reported in the original papers. The main reason for these consistant improvements are due to the per-class NMS of object proposals before sending to relationship head. Also, we found the gap between different methods are also reduced significantly. Our model has similar performance to msdn, while better performance than imp.

Adding RelPN to other Methods

We added our RelPN to various algorithms and compared with the original version.

model relpn bs lr [email protected] [email protected] [email protected] [email protected]
vanilla no 8 5e-3 24.2 10.5 13.8 16.1
vanilla yes 8 5e-3 24.2 12.3 15.8 17.7
imp no 8 5e-3 24.2 16.7 21.7 25.2
imp yes 8 5e-3 24.2 19.2 23.9 26.3
msdn no 8 5e-3 24.2 18.3 23.6 27.1
msdn yes 8 5e-3 24.2 19.2 23.8 26.2

* you can click 'model' in above table to download the checkpoints.

Above, we can see consistant improvements for different algorithms, which demonstrates the effeciveness of our proposed relation proposal network (RelPN).

Also, since much less object pairs (256, originally > 1k) are fed to relation head for predicate classification, the inference time for the models with RelPN is reduced significantly (~2.5 times faster)

Tips and Tricks

Some important observations based on the experiments:

  • Using per-category NMS is important!!!!. We have found that the main reason for the huge gap between the imp-style models and motif-style models is that the later used the per-category nms before sending the graph into the scene graph generator. Will put the quantitative comparison here.

  • Different calculations for frequency prior result in differnt results*. Even change a little bit to the calculation fo frequency prior, the performance of scene graph generation model vary much. In neural motiftnet, we found they turn on filter_non_overlap, filter_empty_rels to filter some triplets and images.

Installation

Prerequisites

  • Python 3.6+
  • Pytorch 1.0
  • CUDA 8.0+

Dependencies

Install all the python dependencies using pip:

pip install -r requirements.txt

and libraries using apt-get:

apt-get update
apt-get install libglib2.0-0
apt-get install libsm6

Data Preparation

  • Visual Genome benchmarking dataset:
Annotations Object Predicate
#Categories 150 50

First, make a folder in the root folder:

mkdir -p datasets/vg_bm

Here, the suffix 'bm' is in short of "benchmark" representing the dataset for benchmarking. We may have other format of vg dataset in the future, e.g., more categories.

Then, download the data and preprocess the data according following this repo. Specifically, after downloading the visual genome dataset, you can follow this guidelines to get the following files:

datasets/vg_bm/imdb_1024.h5
datasets/vg_bm/bbox_distribution.npy
datasets/vg_bm/proposals.h5
datasets/vg_bm/VG-SGG-dicts.json
datasets/vg_bm/VG-SGG.h5

The above files will provide all the data needed for training the object detection models and scene graph generation models listed above.

  • Visual Genome bottom-up and top-down dataset:
Annotations Object Attribute Predicate
#Categories 1600 400 20

Soon, I will add this data loader to train bottom-up and top-down model on more object/predicate/attribute categories.

  • Visual Genome extreme dataset:
Annotations Object Attribute Predicate
#Categories 2500 ~600 ~400

This data loader further increase the number of categories for training more fine-grained visual representations.

Compilation

Compile the cuda dependencies using the following commands:

cd lib/scene_parser/rcnn
python setup.py build develop

After that, you should see all the necessary components, including nms, roi_pool, roi_align are compiled successfully.

Train

Train object detection model:

  • Faster r-cnn model with resnet-101 as backbone:
python main.py --config-file configs/faster_rcnn_res101.yaml

Multi-GPU training:

python -m torch.distributed.launch --nproc_per_node=$NGPUS main.py --config-file configs/faster_rcnn_res101.yaml

where NGPUS is the number of gpus available.

Train scene graph generation model jointly (train detector and sgg as a whole):

  • Vanilla scene graph generation model with resnet-101 as backbone:
python main.py --config-file configs/sgg_res101_joint.yaml --algorithm $ALGORITHM

Multi-GPU training:

python -m torch.distributed.launch --nproc_per_node=$NGPUS main.py --config-file configs/sgg_res101_joint.yaml --algorithm $ALGORITHM

where NGPUS is the number of gpus available. ALGORIHM is the scene graph generation model name.

Train scene graph generation model stepwise (train detector first, and then sgg):

  • Vanilla scene graph generation model with resnet-101 as backbone:
python main.py --config-file configs/sgg_res101_step.yaml --algorithm $ALGORITHM

Multi-GPU training:

python -m torch.distributed.launch --nproc_per_node=$NGPUS main.py --config-file configs/sgg_res101_step.yaml --algorithm $ALGORITHM

where NGPUS is the number of gpus available. ALGORIHM is the scene graph generation model name.

Evaluate

Evaluate object detection model:

  • Faster r-cnn model with resnet-101 as backbone:
python main.py --config-file configs/faster_rcnn_res101.yaml --inference --resume $CHECKPOINT

where CHECKPOINT is the iteration number. By default it will evaluate the whole validation/test set. However, you can specify the number of inference images by appending the following argument:

--inference $YOUR_NUMBER

⚠️ If you want to evaluate the model at your own path, just need to change the MODEL.WEIGHT_DET to your own path in faster_rcnn_res101.yaml.

Evaluate scene graph frequency baseline model:

In this case, you do not need any sgg model checkpoints. To get the evaluation result, object detector is enough. Run the following command:

python main.py --config-file configs/sgg_res101_{joint/step}.yaml --inference --use_freq_prior

In the yaml file, please specify the path MODEL.WEIGHT_DET for your object detector.

Evaluate scene graph generation model:

  • Scene graph generation model with resnet-101 as backbone:
python main.py --config-file configs/sgg_res101_{joint/step}.yaml --inference --resume $CHECKPOINT --algorithm $ALGORITHM
  • Scene graph generation model with resnet-101 as backbone and use frequency prior:
python main.py --config-file configs/sgg_res101_{joint/step}.yaml --inference --resume $CHECKPOINT --algorithm $ALGORITHM --use_freq_prior

Similarly you can also append the ''--inference $YOUR_NUMBER'' to perform partially evaluate.

⚠️ If you want to evaluate the model at your own path, just need to change the MODEL.WEIGHT_SGG to your own path in sgg_res101_{joint/step}.yaml.

Visualization

If you want to visualize some examples, you just simple append the command with:

--visualize

Citation

@inproceedings{yang2018graph,
    title={Graph r-cnn for scene graph generation},
    author={Yang, Jianwei and Lu, Jiasen and Lee, Stefan and Batra, Dhruv and Parikh, Devi},
    booktitle={Proceedings of the European Conference on Computer Vision (ECCV)},
    pages={670--685},
    year={2018}
}

Acknowledgement

We appreciate much the nicely organized code developed by maskrcnn-benchmark. Our codebase is built mostly based on it.

Popular Generation Projects
Popular Scene Projects
Popular Software Development Categories

Get A Weekly Email With Trending Projects For These Categories
No Spam. Unsubscribe easily at any time.
Python
Graph
Neural
Convolutional Neural Networks
Generation
Scene