Jetson cat chasing: Turn on the lawn sprinkler when a neural net sees a cat.
A collection of utilities useful for working with segmented images and the fcn variant of Caffe. See my page for a description of the system.
photon:
cat_sprinkler.cpp Firmware for the photon
python:
count_pascal_mat.py Counts classified pixels in a Pascal Context .mat file
count_pascal.py Coutns classified pixels in a Pascal Context .png file
get_png_palette.py Gets the pallete from a .png file
last10.py Show the last 10 shots from the camera at 1/2 res
mask_out.py Masks out and colorizes the classified pixels in an image
mat2png.py Convert a Pascal Context .mat file to .png
movie.py Spin through a bunch of images on the command line
png2mat.py Convert a Pascal Context .png file to .mat
resize.py Resize an image
seg_fix.py Demo for how to change the segmentation in a file
show_seg.py Overlays displays a base image with the semented pixels
yesno.py Sort a bunch of images into "yes" and "no" buckets
zip_dir.py Zip two directories together
zip_neg.py Zip a directory with a single file
scripts:
cats.sh Simple script to run fcn segmentations on files coming in from FTP
dclean.sh Monitor an FTP directory
tcats.sh The one I use
spon Disable the sprinkler
spoff Enable the sprinkler
sprinkle.sh Start the sprinkler
sprinkle_off.sh Stop the sprinkler
src:
bright.cpp Brighten images
cropper.cpp Crop images to 227 x 227 pixels
example.cpp Simple GPU example used to debug opencv builds
extract_fg.cpp Foreground extraction utility
snapshots.cpp Convert a movie to single frames
fcn: Modified files from the Shellhammer github
infer.py Saves a file in addition to processing a file
batch_infer.py Processes a bunch of files from the command line
tbatch_infer.py The one I use to process inbound cat images
voc-fcn32s:
deploy.prototxt A deployment version of trainval.prototxt