Pretrained weights trained on landmark dataset

[nashory]

Pretrained weights trained on landmark dataset

[rita-qingyu]

Hi, thank you for sharing the trained model. Is this keypoints model trained on the full-version and the finetune model trained on the clean-version?

[PapaMadeleine2022]

@nashory , Hello, when I use your 'pretrained_model/ldmk/pca/pca.h5' to extract dimension-reduced DeLF, it shows error:

...
loaded weights from module "pool" ...
load model from "pretrained_model/ldmk/model/keypoint/ckpt/fix.pth.tar"
load PCA parameters...
Traceback (most recent call last):
  File "extract/extractor.py", line 432, in <module>
    extractor = FeatureExtractor(extractor_config)
  File "extract/extractor.py", line 112, in __init__
    self.pca_mean = h5file['.']['pca_mean'].value
AttributeError: 'Dataset' object has no attribute 'value'

How to fix it? Can you give me some advises?

[PapaMadeleine2022]

@nashory , Hello, when I use your 'pretrained_model/ldmk/pca/pca.h5' to extract dimension-reduced DeLF, it shows error:

...
loaded weights from module "pool" ...
load model from "pretrained_model/ldmk/model/keypoint/ckpt/fix.pth.tar"
load PCA parameters...
Traceback (most recent call last):
  File "extract/extractor.py", line 432, in <module>
    extractor = FeatureExtractor(extractor_config)
  File "extract/extractor.py", line 112, in __init__
    self.pca_mean = h5file['.']['pca_mean'].value
AttributeError: 'Dataset' object has no attribute 'value'

How to fix it? Can you give me some advises?

I use np.array(x) to replace x.value, and now it works.