diff --git a/docs/.nojekyll b/docs/.nojekyll new file mode 100644 index 0000000..e69de29 diff --git a/docs/Logo/Brails.jpg b/docs/Logo/Brails.jpg new file mode 100644 index 0000000..4630b7e Binary files /dev/null and b/docs/Logo/Brails.jpg differ diff --git a/docs/Logo/Logo.pptx b/docs/Logo/Logo.pptx new file mode 100644 index 0000000..b792fb2 Binary files /dev/null and b/docs/Logo/Logo.pptx differ diff --git a/docs/README.md b/docs/README.md new file mode 100644 index 0000000..639133b --- /dev/null +++ b/docs/README.md @@ -0,0 +1,123 @@ +# Building Recognition using AI at Large-Scale. + + + + +## 1. What is BRAILS? + +BRAILS is the acronym for Building Recognition using AI at Large-Scale, +which is an AI-Based pipeline for city-scale building information modeling (BIM). + +## 2. How to use? + + +### a. Data preparation + +Prepare data + + +### b. Train + +#### Roof +##### training +``` +cd src/training/roof/2_train +# train (better to run on a GPU machine) +sh finetune_inception_v3_on_roof_train.sh +``` + +The training takes a long time on laptops. +If you don't want to run the training process, we have a CNN trained on TACC and can be downloaded [here](https://berkeley.box.com/shared/static/awyyc22sjwknn9xg3p7wru4v5zwnlkjp.zip). +Put the downloaded file inside src/training/roof/tmp/roof-traindir/ and unzip it. + +##### testing + + Before your run the test *"sh finetune_inception_v3_on_roof_eval.sh"*, make sure you set the following variables correctly in *"finetune_inception_v3_on_roof_eval.sh"*: + ``` + checkpoint_file = the-path-of-your-checkpoint-file + TEST_DIR = the-path-of-your-testing-images-dir + ``` + We've prepared some labeld images for you to test. These images can be downloaded [here](https://berkeley.box.com/shared/static/wfwf4ku9561lcytldy1p7vkjoobgv9sz.zip). + +Now you can test if the trained CNN works or not: + ``` +cd src/training/roof/2_train +sh finetune_inception_v3_on_roof_eval.sh +``` + + +### c. Predict + +#### Roof +Now we use the CNN to predict roof types based on satellite images. + +Firstly we need to download those images by calling Google API (will cost $). + +``` +cd src/predicting +python downloadRoofImages.py +``` + +To save $, instead of running the above python, you can just download them +[here](https://berkeley.box.com/shared/static/n8l9kusi9eszsnnkefq37fofz22680t2.zip). + + +Now we can make predictions: + +``` +cd src/predicting +sh classifyRoof.sh +``` +This script will look into the BIM file and call CNN to predict the roof type of a building if the image is downloaded. + +If the image is not downloaded, it will assign a null value for the roof type in the new BIM file. + + +### d. Enhance + +Use [*SURF*](https://github.com/charlesxwang/SURF) to predict missing building information: + + +#### Year built + + + + + +#### Number of stories + + + +#### Structure type + + + +#### Occupancy + + + + +## 3. Release of BIM data +SimCenter will post obtained data here. +### City-scale BIM for Atlantic coastal cities, NJ -> [Download](https://berkeley.box.com/shared/static/5tb6gszbbyj35bgpypk1gsdem0ntt5ca.geojson) + + + +## 4. Release of trained CNN +Trained CNNs in the format of Tensorflow checkpoint will be posted here. +### Roof type classifier ->[Download](https://berkeley.box.com/shared/static/awyyc22sjwknn9xg3p7wru4v5zwnlkjp.zip) + + +## 5. How to Cite +You can cite this software as follows: + + + + +## 6. Acknowledgement +This material is based upon work supported by the National Science Foundation under Grant No. 1612843. + +## 7. Contact +Charles Wang, NHERI SimCenter, UC Berkeley, c_w@berkeley.edu + + diff --git a/docs/images/bim-map.png b/docs/images/bim-map.png new file mode 100644 index 0000000..edcda17 Binary files /dev/null and b/docs/images/bim-map.png differ diff --git a/docs/images/occlusion.pptx b/docs/images/occlusion.pptx new file mode 100644 index 0000000..d4fdb07 Binary files /dev/null and b/docs/images/occlusion.pptx differ diff --git a/docs/images/tree.png b/docs/images/tree.png new file mode 100644 index 0000000..ee29e78 Binary files /dev/null and b/docs/images/tree.png differ diff --git a/docs/images/vision-map.png b/docs/images/vision-map.png new file mode 100644 index 0000000..27e4bb5 Binary files /dev/null and b/docs/images/vision-map.png differ diff --git a/docs/index.html b/docs/index.html new file mode 100644 index 0000000..f48a914 --- /dev/null +++ b/docs/index.html @@ -0,0 +1,21 @@ + + +
+ +