README.md

This repository contains the source code of our paper: Wei Yin, Yifan Liu, Chunhua Shen, Anton van den Hengel, Baichuan Sun, The devil is in the labels: Semantic segmentation from sentences

Embedding: https://cloudstor.aarnet.edu.au/plus/s/gXaGsZyvoUwu97t CKPT: https://cloudstor.aarnet.edu.au/plus/s/AtYYaVSVVAlEwve

Quick start:

1. Download the embedding and ckpt and put them into Test_Minist/models/
2. Put your test images in Test_Minist/test_imgs/
3. cd Test_Minist 4 .bash run.sh

See the following branch for a new free text demo: https://github.com/irfanICMLL/SSIW/tree/master

Define the label list

You can choose the label list used for semantic segmentation, for instance:

python tools/test.py --config test_720_ss --user_label dog mouse horse rug_floormat wall person vegetation pizza

Here is the comparison between adding label "cat" or not:

Define new labels

You can also define your own categories described by sentences, for instance:

python tools/test.py --config test_720_ss --new_definitions="{'deer': 'This is an image of deer, similar to sheep or dog.'}"

License

This code is for non-commercial use only. It's released under GPL license. For commercial use, please contact authors.