search

cc-ai / kdb

AI can help Visualizing the Impacts of Climate Change. This is an open forum to share our work
https://mila.quebec/en/ai-society/visualizing-climate-change/
Apache License 2.0
19 stars 3 forks source link

Estimate objects' heights : Explore simulated world depth maps #98

Open melisandeteng opened 5 years ago

melisandeteng commented 5 years ago

The ultimate goal is to train a model to estimate objects' heights in GSV images.

A first subtask is to build a dataset of images from the simulated world with each of the pixels being associated to its real height in meters - the idea is that we can have access to ground truth and a lot of images easily. If we have the camera parameters and the depth map, we can recover the real height of each point in the image (eccluding sky) using euclidian geometry. However, the encoding of the depth map generated in the simulated world is not straightforward. We want to understand what a unit in the depth map corresponds to in cm.

Whether this task is succeeded will determine the strategy for estimating objects' heights in GSV images.

vict0rsch commented 4 years ago

Current results and approach => https://cc-ai.github.io/MUNIT/index.html

@melisandeteng how likely is it that we'll go forward with height estimation?