홍석쓰 블로그

This always causes problems...1) fail to build mmcv from the source, which was recommended by the egohumans repo2) Why don't I just use pip and mim to install mmcv, mmdet as usual. 3) The usual mmcv installation failed to import nms from mmcv.ops import nms4) Ok, I found out that there is something called mmcv-full and I needed that for nms import (and potentially other functions that egohumans ..

Tested methods: WHAM, GVHMR, TRACETL;DRThey all give poor results. They don't have ground estimation. I made issues in each repo to clarify whether I am doing something wrong or it is the fundamental limitation of their methods. (Oct 12th, 2024)https://github.com/yohanshin/WHAM/issues/118 Fail to put multiple persons in the same world frame · Issue #118 · yohanshin/WHAMHI @yohanshin @dalgu90 , T..

top left: rgb video for reference \ top right: rectified slam left video input; right video input is omitted \ bottom left: normalized depth for visualization with a mask where approximates the overlapping regions of left and right camera FOV \ bottom right: normalized predicted disparity from RAFT-stereo 1. Preprocess SLAM left-right camera imageshttps://github.com/hongsukchoi/generic_tools/blo..