Deep Reasoning with Knowledge Graph for Social Relationship Understanding

This repo includes the source code of the paper: "Deep Reasoning with Knowledge Graph for Social Relationship Understanding" (IJCAI 2018) by Zhouxia Wang, Tianshui Chen, Jimmy Ren, Weihao Yu, Hui Cheng, Liang Lin.

Environment

The code is implemented using the Pytorch library with Python 2.7 and has been tested on a desktop with the system of Ubuntu 14.04 LTS.

Dataset

PISC was released by [Li et al. ICCV 2017]. It involves two-level relationship, i.e., coarse-level relationships with 3 categories and fine-level relationships with 6 categories.

PIPA-relation was released by [Sun et al. CVPR 2017]. It covers 5 social domains, which can be further divided into 16 social relationships. On this dataset, we focus on the 16 social relationships.

Models && objects && adjacency matrices

Models, objects and ajacency matrices are in HERE.

Usage

usage: test.py [-h] [-j N] [-b N] [--print-freq N] [--weights PATH]
           [--scale-size SCALE_SIZE] [--world-size WORLD_SIZE] [-n N]
           [--write-out] [--adjacency-matrix PATH] [--crop-size CROP_SIZE]
           [--result-path PATH]
           DIR DIR DIR

PyTorch Relationship

positional arguments:
  DIR                       path to dataset
  DIR                       path to objects (bboxes and categories of objects)
  DIR                       path to test list

optional arguments:
  -h, --help                show this help message and exit
  -j N, --workers N         number of data loading workers (defult: 4)
  -b N, --batch-size N      mini-batch size (default: 1)
  --print-freq N, -p N      print frequency (default: 10)
  --weights PATH            path to weights (default: none)
  --scale-size SCALE_SIZE   input size
  --world-size WORLD_SIZE   number of distributed processes
  -n N, --num-classes N     number of classes / categories
  --write-out               write scores
  --adjacency-matrix PATH   path to adjacency-matrix of graph
  --crop-size CROP_SIZE     crop size
  --result-path PATH        path for saving result (default: none)

Test

Modify the path of data before running the script.

sh test.sh

Result

PISC: Coarse-level

Methods	Intimate	Non-Intimate	No Relation	mAP
Union CNN	72.1	81.8	19.2	58.4
Pair CNN	70.3	80.5	38.8	65.1
Pair CNN + BBox + Union	71.1	81.2	57.9	72.2
Pair CNN + BBox + Global	70.5	80.0	53.7	70.5
Dual-glance	73.1	84.2	59.6	79.7
Ours	81.7	73.4	65.5	82.8

PISC: Fine-level

Methods	Friends	Family	Couple	Professional	Commercial	No Relation	mAP
Union CNN	29.9	58.5	70.7	55.4	43.0	19.6	43.5
Pair CNN	30.2	59.1	69.4	57.5	41.9	34.2	48.2
Pair CNN + BBox + Union	32.5	62.1	73.9	61.4	46.0	52.1	56.9
Pair CNN + BBox + Global	32.2	61.7	72.6	60.8	44.3	51.0	54.6
Dual-glance	35.4	68.1	76.3	70.3	57.6	60.9	63.2
Ours	59.6	64.4	58.6	76.6	39.5	67.7	68.7

PIPA-relation:

Methods	accuracy
Two stream CNN	57.2
Dual-Glance	59.6
Ours	62.3

Citation

@inproceedings{Wang2018Deep,
    title={Deep Reasoning with Knowledge Graph for Social Relationship Understanding},
    author={Zhouxia Wang, Tianshui Chen, Jimmy Ren, Weihao Yu, Hui Cheng, Liang Lin},
    booktitle={International Joint Conference on Artificial Intelligence},
    year={2018}
}

Contributing

For any questions, feel free to open an issue or contact us (zhouzi1212@gmail.com & tianshuichen@gmail.com)

rezacsedu / SR