aledamelio
commented
3 years ago Hi @Morceguito,
thank you for the interest in this work,
the first warning pops out when the face is not fully contained in the frame; as a consequence the face crop in that frame may be wrong. As far as I can understand, probably in one frame the face crop is empty, hence opencv is unable to reshape it. For what concerns the second warning, it is due to the adaptive thresholding procedure used for skin detection. The thresholds are computed as the Highest Density Intervals (HDIs) of the channels color distribution. When more than 2 HDIs are found it just selects the first 2. In general you can ignore this warning.
Anyway we are currently working on an updated version of the package that will fix such issues. The new version will be out soon... so stay tuned!
Alessandro
Morceguito
Hello, I am a Msc student and I am developing a zero-shot learning solution to the face spoofing problem. I am trying to incorporate a rPPG component. I have tried to use a code similar to the one on the basic_demo notebook, only changing it to run through a lot of samples in a loop. It has worked fine on the REPLAY-ATTACK and on OULU but I am having two errors on the datasets CASIA and MSU:
I am using the following params: params = {"video": video, "verb":1, "ROImask":"skin_adapt", "skinAdapt":0.2, "winSize":5}
CASIA: "WARNING! Strange face crop: video frame 36 probably does not contain the whole face... Reshaping Crop" that then leads to "OpenCV(4.1.2) /io/opencv/modules/imgproc/src/resize.cpp:3720: error: (-215:Assertion failed) !ssize.empty() in function 'resize'".
MSU: On the code line "bpmES_chrom, timesES_chrom = chrom.runOffline(**params)" --> ValueError: The length of the input vector x must be greater than padlen, which is 39.
Also, this warning appears a lot of times: "WARNING!! Found more than 2 HDIs in Hue Channel empirical Distribution... Considering only 2" should I be worried?
I would be very thankful for any help.