Access the Challenge Dataset
Training RAW videos:
[Baidu Netdisk]
[OneDrive]
Validation videos (input):
[Baidu Netdisk]
[OneDrive]
Testing videos (input):
[TBD]
[TBD]
The Dataset is for this chanllenge and research purpose only. Commercial use is not permitted.
Content of The Downloaded Package
In the dataset, you may find a folder for training data and a folder for testing data.
Train
In the training dataset folder, you will find all video clips, each of which is named with an index. In the folder, you will find the path like:
"noisy/1(scene number)/100(noise level)/1.tiff(frame number)"
The arrangement of the videos is:
Scene number / noise level / frame number
Test
In the testing dataset, you will find 4 video clip. The same as training dataset, each of the group is named with:
Scene number / noise level / frame number
Note that the submission should be uploaded for each noise level. So make sure you pack the result of each noise level in one folder like:
Noise level / Scene number / frame number
Then, submit each folder of noise level for evaluation.
Read Hyperspectral Cubic Dataset
In the dataset package, you will find a tiff file for each frame. It is in RAW domain captured by a DSLR camera. Here is an example file to read and normalize (given black level 512 and white level 15360) the tiff image:
raw = cv2.imread(path, -1)
# given the path of tiff file
black_level = 512
white_level = 15360
im = raw.astype(np.float32)
im = np.maximum(im - black_level, 0) / (white_level - black_level)
im = np.expand_dims(im, axis=2)
img_shape = im.shape
H = img_shape[0]
W = img_shape[1]
# RGBG
out = np.concatenate((im[0:H:2, 0:W:2, :],
im[0:H:2, 1:W:2, :],
im[1:H:2, 0:W:2, :],
im[1:H:2, 1:W:2, :]), axis=2)
# you can use the out for 4-channel image or the original Bayer pattern 1-channel image