Datasets
Standard Dataset
The supplement training dataset for self-calibrating photometric stereo
- Citation Author(s):
- Submitted by:
- Xi Wang
- Last updated:
- Tue, 05/17/2022 - 22:18
- DOI:
- 10.21227/8mn8-k322
- Research Article Link:
- License:
148 Views- Categories:
Abstract
The PS_Sculpture training dataset introduced by the PS-FCN [1] contains various non-Lambertian reflectances, cast shadows, interreflections and effective noise information. However, for dark materials such as black-phenolic and steel, significant data loss happens due to 8-bit quantification. To lessen this data loss, we design a new supplementary training dataset rendered by 10 blobby objects and 10 other objects freely downloaded from the Internet and the real BRDF data comes from the MERL dataset [2]. Besides, this supplementary dataset includes the SVBRDF and the specific image illuminated by the collocated light. Better deep learning based photometric stereo model could be trained by the combination of the PS_Sculpture dataset and the new supplementary dataset. You could download this dataset at https://pan.baidu.com/s/16WPQhn__-0KlfFfL9VpG0Q, and the code is “misa”.
References:
[1] G. Chen, H. Kai, and K.-Y. K. Wong, "PS-FCN: A Flexible Learning Framework for Photometric Stereo,", in ECCV, 2018.
[2] W. Matusik, H. Pfister, M. Brand, and L. McMillan, "A data-driven reflectance model," in SIGGRAPH, 2003.
The new Bloby dataset contains 25920 scenes and the DetailObject dataset includes 29280 scenes, where half of them is rendered by the single BRDF and another half by the SVBRDF. For every scene, there are 65 images and the first image is the image illuminated by the collocated light. All the images are16-bit format with gamma 1/3 to maintain the reflectance information of dark material. In the training process, the PS_Sculpture_dataset should be read using the source code of PS-FCN [1], and the image in the proposed supplementary dataset should be loaded as origin data and corrected by the gamma 3. The sample code is listed as follows:
img = cv2.imread(img_dir, -1).astype(np.float32)/65535.0
img = np.power(img, 3.0)
References:
[1] G. Chen, H. Kai and K. -Y. K. Wong, "PS-FCN: A Flexible Learning Framework for Photometric Stereo," in ECCV, 2018.
