pifuhd VS VideoPose3D

PIFuHD (Generate 3d Models from a single image) https://github.com/facebookresearch/pifuhd

git clone https://github.com/facebookresearch/pifuhd

Code for https://arxiv.org/abs/2004.00452 found: https://github.com/facebookresearch/pifuhd

I'm very interested in converting images to 3d models. Really cool tech that feels like it's nearly there.

Have you by chance seen any of the work Facebook has been doing related to the 2d to 3d conversions?

https://github.com/facebookresearch/pifuhd

Yes you're correct. That's why I thought having a dome of raspberry pis around a turntable, getting multiple angles at once, but also the movement would provide a solid dataset to be able to train a neural network to generate a 3D model of a person, or an object or possibly even an animal from a single photo, sort of like PifuHD.

PIFuHD

facebookresearch/pifuhd (Python): High-Resolution 3D Human Digitization from A Single Image.

for the pose estimation itself you might want to look at something like this https://github.com/facebookresearch/VideoPose3D

Depending on the chosen algorithm, the HPE system will provide different performance and accuracy results. Let’s see how they correlate in terms of our experiment with two of the most popular human pose estimation models, VideoPose3D and BlazePose.

One thing you can start with is just extracting poses from fencing matches. There are libraries out there that your can download and implement, here's one that popped from a quick search: https://github.com/facebookresearch/VideoPose3D (figure skating example)

Source