Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
AngelWings1997 601c2fb5c5 | 1 year ago | |
---|---|---|
data | 1 year ago | |
datasets | 1 year ago | |
images | 4 years ago | |
networks | 4 years ago | |
pretrain_model/hub/checkpoints | 1 year ago | |
utils | 1 year ago | |
.gitignore | 1 year ago | |
README.md | 2 years ago | |
checkpoint_first_test.pth.tar | 1 year ago | |
element_wise_layer.py | 4 years ago | |
first_test_score | 1 year ago | |
ggnn.py | 1 year ago | |
main.py | 1 year ago | |
main_coco.sh | 4 years ago | |
main_vg.sh | 4 years ago | |
main_voc07.sh | 1 year ago | |
main_voc12.sh | 4 years ago | |
model_best_first_test.pth.tar | 1 year ago | |
models.py | 4 years ago | |
semantic.py | 4 years ago | |
train_voc2007.py | 1 year ago | |
train_voc2007.sh | 1 year ago |
Implementation of the paper: "Learning Semantic-Specific Graph Representation for Multi-Label Image Recognition" (ICCV 2019) by Tianshui Chen, Muxin Xu, Xiaolu Hui, Hefeng Wu, Liang Lin.
Python 2.7
Pytorch 0.4.1
Ubuntu 14.04 LTS
Microsoft COCO - 80 common object categories
Pascal VOC 2007 - 20 common object categories
Pascal VOC 2012 - 20 common object categories
VisualGenome - subset of VG, covering 500 most common object categories
You can download the data files and our best models here
password: ep6u
git clone https://github.com/Mu-xsan/SSGRL.git
cd SSGRL
mkdir data (download the data needed and put here)
bash main_coco.sh [GPU_id] [Remark for this experiment]
bash main_voc07.sh [GPU_id] [Remark for this experiment]
bash main_voc12.sh [GPU_id] [Remark for this experiment]
bash main_vg.sh [GPU_id] [Remark for this experiment]
Microsoft COCO:
Method | mAP | CP | CR | CF1 | OP | OR | OF1 |
---|---|---|---|---|---|---|---|
SSGRL | 83.8 | 89.9 | 68.5 | 76.8 | 91.3 | 70.8 | 79.7 |
Pascal VOC 2007:
Classes | AP(SSGRL) | AP(pre) |
---|---|---|
aeroplane | 99.5 | 99.7 |
bicycle | 97.1 | 98.4 |
bird | 97.6 | 98.0 |
boat | 97.8 | 97.6 |
bottle | 82.6 | 85.7 |
bus | 94.8 | 96.2 |
car | 96.7 | 98.2 |
cat | 98.1 | 98.8 |
chair | 78.0 | 82.0 |
cow | 97.0 | 98.1 |
diningtable | 85.6 | 89.7 |
dog | 97.8 | 98.8 |
horse | 98.3 | 98.7 |
motorbike | 96.4 | 97.0 |
person | 98.8 | 99.0 |
pottedplant | 84.9 | 86.9 |
sheep | 96.5 | 98.1 |
sofa | 79.8 | 85.8 |
train | 98.4 | 99.0 |
tvmonitor | 92.8 | 93.7 |
mAP | 93.4 | 95.0 |
Pascal VOC 2012:
Classes | AP(SSGRL) | AP(pre) |
---|---|---|
aeroplane | 99.5 | 99.7 |
bicycle | 95.1 | 96.1 |
bird | 97.4 | 97.7 |
boat | 96.4 | 96.5 |
bottle | 85.8 | 86.9 |
bus | 94.5 | 95.8 |
car | 93.7 | 95.0 |
cat | 98.9 | 98.9 |
chair | 86.7 | 88.3 |
cow | 96.3 | 97.6 |
diningtable | 84.6 | 87.4 |
dog | 98.9 | 99.1 |
horse | 98.6 | 99.2 |
motorbike | 96.2 | 97.3 |
person | 98.7 | 99.0 |
pottedplant | 82.2 | 84.8 |
sheep | 98.2 | 98.3 |
sofa | 84.2 | 85.8 |
train | 98.1 | 99.2 |
tvmonitor | 93.5 | 94.1 |
mAP | 93.9 | 94.8 |
VisualGenome-500
Method | mAP |
---|---|
SSGRL | 36.6 |
@inproceedings{chen2019learning,
title={Learning semantic-specific graph representation for multi-label image recognition},
author={Chen, Tianshui and Xu, Muxin and Hui, Xiaolu and Wu, Hefeng and Lin, Liang},
booktitle={Proceedings of the IEEE International Conference on Computer Vision},
pages={522--531},
year={2019}
}
@article{chen2020knowledge,
title={Knowledge-guided multi-label few-shot learning for general image recognition},
author={Chen, Tianshui and Lin, Liang and Hui, Xiaolu and Chen, Riquan and Wu, Hefeng},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
year={2022},
publisher={IEEE}
}
For any questions, feel free to open an issue or contact us (tianshuichen@gmail.com & xumx7@mail2.sysu.edu.cn & huixlu@mail2.sysu.edu.cn)
考虑语义标签与图像内容的多标签分类
Text Python Shell
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》