Anonymize

temporarily giving up, halfway done rewriting my anonymize.py but better and for onnx, see the other file

trying to start video processing, but limited success so far, unfortunately

tried it again, one fewer false positive, but not great results, honestly

increased number of accurate detections and improved accuracy. major changes to the facial detection.

possible next step: using a different model, maybe using onnx centerface? starting to implement additional functions, too

updated processing of image and validation of faces. broke down the parts of the face compared to previously. Unfortunately, I got rid of one false positive and made a new one lol. I think I might need to revisit my strategy

I've decided to redo the validation and process image functions since there's way too many false positives compared to the rate of detection, although I'm still proud of the detection. the debug files are really good though, I'm proud of myself for that idea since I can see the detection specifically. Not as much improvement as I'd hoped 9 hours in :(

beefed up my detection using more math, different scales, and varying thresholds. we finally detect some of the faces in the blurry, small crowd picture

updated the pixel and block features, this is what the blocks look like. you can see a little bit of false catching, and missing the side of someone's face

improving the facial validation! tested out new parameters and multiple levels of checks, also peep what the pixel version looks like. I think the issue was low quality images before. It took a hot minute for me to figure out some of this. I also

another version of detecting faces, working on improvements, better accuracy for detected faces, but could still use work

updated a couple of functions, tried a few more runs, overall slightly worse and not better. I might pull out the cnn again and go from there

testing repeatedly, we're making progress, but it's not great, able to get some of the faces, but no where near all of them so far

made a list of features and typed up a first iteration with a few features, here's what I hope it'll look like when I'm done