Hey I found a issue about the colour space. When detect the input video, the opencv will read the frames in BGR colour space, however they are not converted into RGB format and then directly input into the detect image function, where the PIL treat them as RGB pictures. This could affect the accuracy when it is recognizing the colour related objects (e.g. Traffic Lights). Hope the originator can improve it.
Hey I found a issue about the colour space. When detect the input video, the opencv will read the frames in BGR colour space, however they are not converted into RGB format and then directly input into the detect image function, where the PIL treat them as RGB pictures. This could affect the accuracy when it is recognizing the colour related objects (e.g. Traffic Lights). Hope the originator can improve it.