running with an error

[knazeri]

running with an error

[Haoyanlong]

Can you load the pretrained model in places2, I have a bug when load the pretrained model.

[yuxiaoleipeng]

Can you load the pretrained model in places2, I have a bug when load the pretrained model.

yes, do you have qq or wechart? but i can't load the image data

[knazeri]

@yuxiaoleipeng Our model is trained using an extra irregular mask dataset which you need to download. You do not need edge dataset unless you want to train the model using external edge information. Please follow my comment here to download and create mask dataset:
#11 (comment)

[Haoyanlong]

Can you load the pretrained model in places2, I have a bug when load the pretrained model.

yes, do you have qq or wechart? but i can't load the image data

@yuxiaoleipeng The input data of network contains three iterms, the image with mask , the edge image of original image(gained by edge detector(cv2.Canny)) and mask image when you train the model. You must generate edge image of original image by Canny before training.

[knazeri]

@Haoyanlong If you consider Canny as the source of edge information, our model automatically extracts them from the training data using skimage canny edge detector:

edge-connect/src/dataset.py

Line 98 in 6ec17e5

return canny(img, sigma=sigma, mask=mask).astype(np.float)

If you want to explore other edge detecting systems, then you need to generate them before training.

[Haoyanlong]

OK, thank you very much!I see.