src-apps

What is the src-apps directory?

These are a series of very simple single-purpose apps to show how to use the DarkHelp C++ API. These are meant to be used as a starting point for people who want to use Darknet and DarkHelp. They are not meant to be final production or commercial-ready applications.

Find the one that seems to do what you want, and modify it as needed. The examples as shown below are typically done with the Rolodex example dataset.

When running these example apps on Windows, note the DLL files in the C:\Program Files\Darknet\bin\ and C:\Program Files\DarkHelp\bin\ directories. The examples in this directory will need access to the same dependencies (OpenCV, Darknet, possibly CUDA and cuDNN, etc).

App Descriptions

• process_single_image.cpp Loads a neural network, call predict() on a single image, and display the results on the console. Should look similar to this:
  • Run: src-apps/process_single_image Rolodex.cfg Rolodex.names Rolodex_best.weights page_70.png
  • Output should look similar to this:

page_70.png prediction results: 10
-> 1/10: "user level 2 100%" #7 prob=0.996683 x=236 y=421 w=15 h=23 oid=0 tile=0 entries=1
-> 2/10: "date of birth 100%" #2 prob=0.999682 x=143 y=331 w=207 h=23 oid=0 tile=0 entries=1
-> 3/10: "female 100%" #4 prob=0.999683 x=535 y=331 w=13 h=22 oid=0 tile=0 entries=1
-> 4/10: "marital status 100%" #5 prob=0.999713 x=299 y=371 w=16 h=23 oid=0 tile=0 entries=1
-> 5/10: "serial number 100%" #12 prob=0.999727 x=423 y=461 w=219 h=23 oid=0 tile=0 entries=1
-> 6/10: "address 100%" #1 prob=0.999777 x=50 y=240 w=437 h=67 oid=0 tile=0 entries=1
-> 7/10: "avt 100%" #11 prob=0.999786 x=533 y=420 w=208 h=24 oid=0 tile=0 entries=1
-> 8/10: "barcode 100%" #14 prob=0.999866 x=548 y=523 w=186 h=92 oid=0 tile=0 entries=1
-> 9/10: "name 100%" #0 prob=0.999887 x=60 y=176 w=430 h=42 oid=0 tile=0 entries=1
-> 10/10: "qr code 100%" #13 prob=0.999931 x=98 y=522 w=101 h=100 oid=0 tile=0 entries=1

• display_single_image.cpp Loads a neural network, call predict() on a single image, annotate the image with the results, and display using a GUI window.

  • Run: src-apps/display_single_image Rolodex.cfg Rolodex.names Rolodex_best.weights page_70.png

• display_single_image_snapping.cpp Similar to the previous app, but enables the "snapping" feature.

  • Run: src-apps/display_single_image_snapping Rolodex.cfg Rolodex.names Rolodex_best.weights page_70.png

• display_single_image_custom_settings.cpp Similiar to the previous app, shows how to customize a number of the DarkHelp and Darknet settings.

  • Run: src-apps/display_single_image_custom_settings Rolodex.cfg Rolodex.names Rolodex_best.weights page_70.png

• process_video_webcam.cpp Loads a neural network, open the first webcam, and continuously apply the network to every video frame. Press ESC to exit.

  • If you have a webcam, try and run the MSCOCO pre-trained weights with this one. Download the MSCOCO YOLOv4-tiny weights.
  • Run: src-apps/process_video_webcam yolov4-tiny.cfg coco.names yolov4-tiny.weights

• save_webcam_to_video.cpp Similar to the previous app, but the output is saved back out to disk, named out.mp4. Press ESC to stop recording and exit.

  • Run: src-apps/save_webcam_to_video yolov4-tiny.cfg coco.names yolov4-tiny.weights

• rotate_images.cpp Detects image rotation, rotates the images back to level, and then re-runs the neural network to get more accurate detections.

  • See: https://www.ccoderun.ca/programming/2023-11-26_YOLO_and_image_rotation/
  • Run: src-apps/rotate_image Perfs.cfg Perfs.names Perfs_best.weights ~/nn/Perfs/set_01/film*.jpg

• video_display_realtime.cpp Loads a video file and displays it at the exact speed it was recorded, running the neural network on each frame.

  • Run: src-apps/video_display_realtime animals.cfg animals.names animals_best.weights zoo.m4v

• video_object_counter.cpp Combines predictions and object tracker to count the number of objects.

  • See: output should be similar to this: https://youtu.be/2biQpVRFhbk
  • Run: src-apps/video_object_counter pigs.cfg pigs.names pigs_best.weights farm.m4v

• using_c_api.cpp Demonstrates how to use the C API.

  • Run: src-apps/using_c_api Rolodex.cfg Rolodex.names Rolodex_best.weights page_70.png

• process_many_images_on_threads.cpp Demonstrates how to use the DHThreads class to load multiple copies of a neural network to process many image files at once.

  • Run: src-apps/process_many_images_on_threads yolov4-tiny.cfg coco.names yolov4-tiny.weights images/

• display_using_bundle.cpp Demonstrates how to use a .dh combined bundle file with DarkHelp. Run DarkHelpCombine to create the .dh file. Compare this code against display_single_image.cpp.

  • Run: DarkHelpCombine password Rolodex.cfg Rolodex.names Rolodex_best.weights followed by display_using_bundle Rolodex.dh password page_70.png

• resize_corners.cpp Looks for classes named "TL", "TR", "BR", "BL" and resizes those annotations to a fixed size.

  • Run: src-apps/resize_corners ~/nn/animals/animals.names