initial commit

Li-Chengyang · Nov 9, 2018 · dd7937d · dd7937d
1 parent 45a3cc5
commit dd7937d
Show file tree

Hide file tree

Showing 64 changed files with 35,862 additions and 2 deletions.
diff --git a/LICENSE b/LICENSE
@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2018 Chengyang Li
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
diff --git a/README.md b/README.md
@@ -1,2 +1,72 @@
-# MSDS-RCNN
-''Multispectral Pedestrian Detection via Simultaneous Detection and Segmentation''. BMVC 2018.
+### Multispectral Pedestrian Detection via Simultaneous Detection and Segmentation
+Edited by Chengyang Li, Zhejiang University.
+
+Demo code of our paper [Multispectral Pedestrian Detection via Simultaneous Detection and Segmentation](https://arxiv.org/abs/1808.04818) by Chengyang Li, Dan Song, Ruofeng Tong and Min Tang. BMVC 2018. [[project link]](https://li-chengyang.github.io/home/MSDS-RCNN/).
+
+<img src="figures/overview.png" width="800px" height="400px"/>
+
+### Demo
+0. Prerequisites
+  Basic Tensorflow and Python package installation.
+  This code is tested on Ubuntu14.04, tf1.2, Python2.7 and Ubuntu16.04, tf1.11, Python3.5.
+
+1. Clone the repository
+  ```Shell
+  git clone https://github.com/Li-Chengyang/MSDS-RCNN.git
+  ```
+
+2. Update your -arch in setup script to match your GPU
+  ```Shell
+  cd msds-rcnn/lib
+  # Change the GPU architecture (-arch) if necessary
+  vim setup.py
+  ```
+
+3. Build the Cython modules
+  ```Shell
+  make clean
+  make
+  cd ..
+  ```
+
+4. Download the pre-trained model
+[One driver](https://1drv.ms/u/s!AtMRVLTL5T5eb5_kGuk3AZxDT4o) trained on KAIST using original training annotaions.
+Untar files to output/vgg16/.
+
+5. Run demo 
+  ```Shell
+  python tools/demo.py
+  ```
+
+### Detection performance
+
+<img src="figures/comparisons.png" width="800px" height="250px"/>
+
+**Note**: 
+Since the original annotations of the test set contain many problematic bounding boxes, we use the [improved testing annotations](http://paul.rutgers.edu/%7Ejl1322/multispectral.htm) provided by Liu et al. to enable a reliable comparison.
+
+### Downloads
+
+[Human baseline](https://drive.google.com/open?id=1hNLSRPpQWRANf62kG58X6dI4uIMKwL3n)
+
+[Sanitized training annotations](https://goo.gl/forms/Lfgd3vcx4ZrFxs0J3)
+
+[Detection results](https://drive.google.com/open?id=1MLejnwZr7C1imUa9emyVJiUH5CxbYw-T)
+
+### Acknowledgements
+
+Thanks to Xinlei Chen, this pipeline is largely built on his example tensorflow Faster R-CNN code available at:
+[https://github.com/endernewton/tf-faster-rcnn](https://github.com/endernewton/tf-faster-rcnn)
+
+### Citing our paper
+If you find our work useful in your research, please consider citing:
+
+```
+@article{li2018multispectral,
+  title={Multispectral Pedestrian Detection via Simultaneous Detection and Segmentation},
+  author={Li, Chengyang and Song, Dan and Tong, Ruofeng and Tang, Min},
+  journal={arXiv preprint arXiv:1808.04818},
+  year={2018}
+}
+```
+
diff --git a/data/demo/set06_V001_I00459_lwir.png b/data/demo/set06_V001_I00459_lwir.png
diff --git a/data/demo/set06_V001_I00459_visible.png b/data/demo/set06_V001_I00459_visible.png
diff --git a/data/demo/set08_V002_I01419_lwir.png b/data/demo/set08_V002_I01419_lwir.png
diff --git a/data/demo/set08_V002_I01419_visible.png b/data/demo/set08_V002_I01419_visible.png
diff --git a/data/demo/set09_V000_I01139_lwir.png b/data/demo/set09_V000_I01139_lwir.png
diff --git a/data/demo/set09_V000_I01139_visible.png b/data/demo/set09_V000_I01139_visible.png
diff --git a/data/demo/set09_V000_I01959_lwir.png b/data/demo/set09_V000_I01959_lwir.png
diff --git a/data/demo/set09_V000_I01959_visible.png b/data/demo/set09_V000_I01959_visible.png
diff --git a/data/demo/set10_V001_I03279_lwir.png b/data/demo/set10_V001_I03279_lwir.png
diff --git a/data/demo/set10_V001_I03279_visible.png b/data/demo/set10_V001_I03279_visible.png
diff --git a/data/demo/set11_V000_I00159_lwir.png b/data/demo/set11_V000_I00159_lwir.png
diff --git a/data/demo/set11_V000_I00159_visible.png b/data/demo/set11_V000_I00159_visible.png
diff --git a/figures/comparisons.png b/figures/comparisons.png
diff --git a/figures/highlight.png b/figures/highlight.png
diff --git a/figures/overview.png b/figures/overview.png
diff --git a/lib/Makefile b/lib/Makefile
@@ -0,0 +1,6 @@
+all:
+	python setup.py build_ext --inplace
+	rm -rf build
+clean:
+	rm -rf */*.pyc
+	rm -rf */*.so
diff --git a/lib/__init__.py b/lib/__init__.py
diff --git a/lib/datasets/KAISTdevkit-matlab-wrapper/README.md b/lib/datasets/KAISTdevkit-matlab-wrapper/README.md
@@ -0,0 +1,7 @@
+## matlab wrapper for kaist evaluation
+
+Written by Chengyang Li based on the original demo code from Soonmin Hwang.
+
+See demo_test.m for an example.
+
+**Note**: We provide evalutaion results in terms of log-average miss rate as well as recall on all 9 different settings, i.e. Reasonable-all, Reasonable-day, Reasonable-night, Scale=near, Scale=medium, Scale=far, Occ=none, Occ=partial, Occ=heavy (see CVPR15 paper for details). Both using the origninal and the improved (in parentheses) test annotaions. However we strongly suggest you report results using the improved test annotaions only to enable a fair comparision.