A quick way to practice object detection on images and videos for common classes using TensorFlow. Fully operational by any user without Machine Learning experience.
imageio
imageio-ffmpeg (for videos)
tensorflow
tqdm
pip install tflash
Note: due to the incompatibility of numpy versions, in some cases tensorflow needs to be (re-)installed after installing imageio
import tflash
flash = tflash.Detector()
result = flash.detect("a_pic.jpg", print_output="a_result.jpg") # can be set to False
result = flash.detect("mypic.jpg", min_score=0.75) # default: 0.5
my_pics = ["pic001.jpg", "pic002.jpg", "pic004.jpg"]
result = flash.detect_multiple(my_pics)
Output:
- Dict of detection results
- Labeled image(s)
detections = result["detections"] # formatted as dict
print("Saved to {}".format(result["output"]) # output filename
Alter font (default is Roboto size 20):
flash.set_font("arial.ttf")
flash.set_font_size(12)
flash.set_font("dosis.ttf", size=15)
Some good ones are provided at Tensorflow model zoo: https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/tf1_detection_zoo.md
Download a file with output "Boxes" from the link above, extract and use only the one with *.pb extension, e.g., frozen_inference_graph.pb
Load in TFlash using the command:
flash.load_model("frozen_inference_graph.pb")
or
flash = tflash.Detector("any_other_model_file.pb")