You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hello @natanielruiz
I have a few doubts that I would like to clarify.
Is this model suitable for estimating head pose during low-lit and night vision conditions?
If yes, is it expected to give similar accuracy as normal well-lit conditions?
Also, are we supposed to include some changes for it to work in dim-lit conditions? If yes, could you please mention them?
Hello @natanielruiz
I have a few doubts that I would like to clarify.
Is this model suitable for estimating head pose during low-lit and night vision conditions?
If yes, is it expected to give similar accuracy as normal well-lit conditions?
Also, are we supposed to include some changes for it to work in dim-lit conditions? If yes, could you please mention them?
In the code file the following snippet is used:
transformations = transforms.Compose([transforms.Scale(224), transforms.CenterCrop(224), transforms.ToTensor(), transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])])
Is there any particular need to transform the image into 224 X 224?
If yes, could you please explain the need to do the same?
P.S. I am really grateful this implementation exists.
The text was updated successfully, but these errors were encountered: