image problem about sc-sfmlearner-release HOT 5 CLOSED

ValueError:Image of type float must be between -1 and 1

from sc-sfmlearner-release.

It says that the the "imresize" function requires that the image of type float must be between -1 and 1. Here, the image is first loaded in "sequence_floder.py" as float type by
"
def load_as_float(path):
return imread(path).astype(np.float32)
"

Can you try to change it to
"
def load_as_float(path):
return imread(path)
"
and have a try?

from sc-sfmlearner-release.

Or you can try to change the line76 in custom_transforms.py

scaled_images = [imresize(im, (scaled_h, scaled_w)) for im in images]

to

scaled_images = [np.array(Image.fromarray(im.astype(np.uint8)).resize((scaled_w, scaled_h))).astype(np.float32) for im in images]

Also, do not forget add "from PIL import Image" before using Image.fromarray()

from sc-sfmlearner-release.

Hi,
I'm very happy to have received your response.This change worked.
scaled_images = [np.array(Image.fromarray(im.astype(np.uint8)).resize((scaled_w,
scaled_h))).astype(np.float32) for im in images]
In the subsequent training, a problem occurred in inversed_warp.py (typeerror:grid_sample got an
unexpected keyword argument 'align_corners'). Normal training after deletion this.
Thank you.

from sc-sfmlearner-release.

It is due to pytorch version.
"align_corners" is a new keyword in pytorch after 1.4 (or 1.3 that I forgot).
You can update pytorch to the latest version, or you can just delete 'align_corners' in grid_sample function.

from sc-sfmlearner-release.