csxmli2016 / dmdnet Goto Github PK

dmdnet still need dictionaries file, several G?

Please provide a way to convert to ONNX or make one available.

All instructions are executed, but the result is not saved, what could be the matter?

CUDA_VISIBLE_DEVICES=0 python main_test.py -s ./TestExamples/TestGenericRestoration/GenericLists.txt -d ./TestExamples/Results_TestGenericRestoration
Using device : cuda
Model params : 40.36M
Restoring test_face_orig.png with only generic restoration
There may be something wrong with the detected component locations. Continue...

File "/usr/local/lib/python3.9/site-packages/torch/nn/functional.py", line 3959, in interpolate
return torch._C._nn.upsample_bilinear2d(input, output_size, align_corners, scale_factors)
RuntimeError: Input and output sizes should be greater than 0, but got input (H: 0, W: 0) output (H: 40, W: 40)

Good job on this enhancer; I'm wondering why it isn't more known...

While experimenting with it, I'm getting this error sometimes:

I couldn't find out where this comes from, is it because of different image dimensions or missing landmarks?

Hello,sorry to bother you.I want to know the details about how to accomplish three stages Initialization,Forward Update,Backward Update during training DMDNet.
Before training DMDNet,should I do something about Generic Feature Extractor to finish Initialization stage

This is great work! Could you release the whole code? Thank you!

Img = cv2.resize(Img, (512,512), interpolation = cv2.INTER_AREA)
This is how it should be, cause current code is limiting bigger images than 512 res from processing which is very weird, not all 512 res image are actual 512 res
updated func

def read_img_tensor(img_path=None, return_landmark=True): #rgb -1~1
Img = cv2.imread(img_path, cv2.IMREAD_UNCHANGED) # BGR or G
if Img.ndim == 2:
Img = cv2.cvtColor(Img, cv2.COLOR_GRAY2RGB) # GGG
else:
Img = cv2.cvtColor(Img, cv2.COLOR_BGR2RGB) # RGB

Img = cv2.resize(Img, (512,512), interpolation = cv2.INTER_AREA)

ImgForLands = Img.copy()
Img = Img.transpose((2, 0, 1))/255.0
Img = torch.from_numpy(Img).float()
normalize(Img, [0.5,0.5,0.5], [0.5,0.5,0.5], inplace=True)
ImgTensor = Img.unsqueeze(0)
SelectPred = None
if return_landmark:
    try:
        PredsAll = FaceDetection.get_landmarks(ImgForLands)
    except:
        print('Error in detecting this face {}. Continue...'.format(img_path))
    if PredsAll is None:
        print('Warning: No face is detected in {}. Continue...'.format(img_path))
        return ImgTensor, None
    ins = 0
    if len(PredsAll)!=1:
        hights = []
        for l in PredsAll:
            hights.append(l[8,1] - l[19,1])
        ins = hights.index(max(hights))
        print('Warning: Too many faces are detected, only handle the largest one...')
    SelectPred = PredsAll[ins]
return ImgTensor, SelectPred

I tried to mix up different faces like matt damon face from 0009 onto women faces, it did 0 difference to the image, if it would work, it would be his facial features on woman's face , am i right ?