Reading through the main page "CCTV3
Frigate compatible model. Only trained on gray images. Can only detect person. Resolution changed from 320x320 to 340x340. Achieves the same accuracy on both gray and color images. Approximately 18% slower inference time."
So looks like we get same accuracy as default model at the cost of inference time?
And maybe adding the instruction from this request to the main page would help
Reading through the main page "CCTV3
Frigate compatible model. Only trained on gray images. Can only detect person. Resolution changed from 320x320 to 340x340. Achieves the same accuracy on both gray and color images. Approximately 18% slower inference time."
So looks like we get same accuracy as default model at the cost of inference time?
And maybe adding the instruction from this request to the main page would help
https://codeberg.org/Curid/TF-CCTV/issues/72