lib/dataset: dataset files we used in training
lib/networks: networks for each pipeline
lib/models: pipelins including the loss function
lib/train*: training code
lib/configs: training configs
This needs 3D asset from: https://www.gobotree.com/cat/3d-people [detail asset can be found at gobotree_filename.txt] Also https://www.3dscanstore.com/archviz-3d-models/female-archviz-3d-models [from 1 to 22] The 3D asset is not expensive, and easy to use.
lib/render/data: files about hdr and alignment vectors.
lib/render/ShadowPipeline: This can automatically generate maya file for relighting dataset.
lib/render/AutoRender: This can do batch render to generate dataset.
Render Spec we are using: Maya 2019 and Arnold Render 5.3(MtoA 3.2.0) To accelerate the render speed, we highly recommend to enable GPU-support.
To create maya file, please set the file paths in lib/render/ShadowPipeline.py.
And then in Maya command line run:
python lib/render/ShadowPipeline.py
After creating maya file, to do batch render the image, please run.
python lib/render/AutoRender.py
Due to the current license constrain, https://www.gobotree.com/acceptable/, we can not share the 3D asset to you. After you get the 3D asset from gobtree, please contact me to access the rendered images.