judyxyang / deep-hdr-with-pytorch Goto Github PK

This project is an attempt to reproduce the experiment described here [1]. It consist in merging three LDR images of different exposition of the same scene into one HDR image. To do so, an optical flow algorithm is used to align the three images. Then, a CNN with 4 layers is used to model the merging process. You can download the low quality images report here or download the high quality images here. Both reports are written in french. The resulting HDR images of 15 scenes generated by our CNN and the results of the original paper can be downloaded here.

[1] N. K. Kalantari and R. Ramamoorthi, “Deep high dynamic range imaging of dynamic scenes,” ACM Transactions on Graphics (Proceedings of SIGGRAPH 2017), vol. 36, no. 4, 2017.

Scene folder need to follow this structure:

---scene
      |---test
      |---training

The path to the root scene folder can be changed in the file Constants.py