Can you explain the dataset details? When I checked the track2_train, it has the same image name with _ N, _NE,_NW,_S,_SE,_SW and W.
What would be the model input and output? could you explain the image name and what it the expected model outputs?
Posted by: sabarinathan @ May 17, 2020, 12:28 a.m.Hi,
The details can be found in the "Get Data" section, under "Participate".
In the training dataset of Track2 and Track3, each scene is rendered with 40 different illumination settings (5 color temperatures {2500K, 3500K, 4500K, 5500K, 6500K} and 8 directions {N, NE, E, SE, S, SW, W, NW}), the file name is thus Image{imageid}_{color_temperature}_{direction}.png .
In the validation or test phase of Track 2, you are given some images and you are asked to predict the color temperature and direction of each image. The format of the output is also explained in the "Evaluation" section (temperatures are in [0,1] and angle directions in [0,360], with 0 being North and 90 being East).
In the validation or test phase of Track 3, you are given pairs of images. Each pair contains an input image and a guide image. The input image was rendered in one of the 40 illumination settings, the guide image was also rendered in one of the 40 illumination settings. Your goal is to relight the input image to make it have the same illumination setting as those of the guide image, so your output "target image" should have the content of the input image and the illumination settings of the guide image.
Hope this answers your question, and please let us know if there is still something unclear.
Posted by: Majed @ May 17, 2020, 9:47 a.m.