search

FilippoAleotti / mobilePydnet

Pydnet on mobile devices
Apache License 2.0
253 stars 40 forks source link

How did you pick the images from Microsoft COCO and OpenImages? #13

Closed yan99033 closed 4 years ago

yan99033 commented 4 years ago

I have a follow-up question about issue #11: In your paper "Real-time single image depth perception in the wild with handheld devices", you mentioned about picking 447k images from the two datasets. Did you systematically select a fixed amount of images across different categories? What's the strategy?

Are you planning to release the WILD dataset as well as the training code?

Great work, by the way.

FilippoAleotti commented 4 years ago

Hi, For the training, I did not apply any content-based strategy, but you can exploit object annotations provided by the datasets to make a content-specific dataset. On the contrary, with the aim of being robust in the wild, I just applied checks in terms of shapes (no portrait images) and I also discarded those images with a black border (if the sum of all the pixels belonging to the first or last 50 columns picture is very low then I skip the image). This means that some images are really hard to explain for a monocular network, but that’s fine because I don’t know which images the network will see at testing time (think about this) I do not plan to release the dataset, but just the training code and the list of images in the future.

yan99033 commented 4 years ago

Thanks for the reply. Looking forward to the release of the training code and the list of images.

May I know which links you used to download the datasets? For Microsoft COCO, is it "2017 Unlabelled images"? image

The OpenImages dataset is huge (18TB in total), which one did you use?

FilippoAleotti commented 4 years ago

Hi ,sorry for the delay in responding you. I've added the image list in single_inference folder.