Adam W. Harley
Adam W. Harley
Great question! This is a re-implementation of a specific neural network from "Gradient-based learning applied to document recognition" by Lecun, Bottou, Bengio, and Haffner (1998), and the connection pattern was...
Thanks, this does look useful! That is indeed a slow part of the code. I'll keep this issue open until I integrate the idea of your snippet.
You can swap TwoImageData with ImageSegData in the prototxt. See the "res" example to see how to use ImageSegData. I didn't mean to close the first issue before actually fixing...
Yes: I recommend pre-training the embeddings using VOC 2012, or COCO. You can also train them end-to-end in the full pipeline. I provided a config file for an embedding network.
If you want a dense flow map, I think it would be better to use [RAFT](https://github.com/princeton-vl/RAFT) or [GMFlow](https://github.com/haofeixu/gmflow).
Hi, sorry for the delay. I'm aiming for this Friday.
I'm not sure I get it. Isn't there only one `requirements.txt` file?
`z` should consistently be depth, not height. `z` is zeros there to match the depth axis unsqueezed from `rgb_camB`.
Wow thanks. Good catch.
In that context, `camB` is the camera where we captured the image, and `camA` is the camera which defines the 3D/BEV grid. (The 3D grid faces the same way as...