Created by Victor Basu.
Show this thread
New tutorial on http://keras.io : depth estimation. Given an image, predict the distance of each point to the camera. Important task for reconstructing scene geometry from images!https://keras.io/examples/vision/depth_estimation/ …
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Great tutorial!
But the results in the end there look pretty bad.
I recommend training on KITTY 2015
The problem domain (road etc) is more constrained and better suited for such an underpowered model, and modest training goals (e.g number of epochs) -
The results in these tutorials are always low because they're live notebooks that we have to run on limited resources. You always have to increase the epoch count or switch to a larger dataset to get the real results. Instructions are specified
End of conversation
New conversation -
-
-
Also, there's my Android implementation for monocular depth estimation, if someone is looking for an app ->https://github.com/shubham0204/Realtime_MiDaS_Depth_Estimation_Android …
-
It uses the MiDaS model available on TensorFlow Hub ->https://tfhub.dev/intel/midas/v2_1_small/1 …
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.