A pytorch CSRNET implementation for crowd counting from images.
Current naming support shanghaitech dataset.
In order to run:
-
put the shanghai dataset under /datasets folder in the root path of the repo
-
Set absolute paths in presets.py
-
preprocess density maps by running prepare_dataset.py
-
run train.py script
-
evaluate your model with eval.py
With the example preset you can train a sub-70 MAE on ShanghaiTech_A_test.