Closed k-sokolov closed 4 years ago
Training a custom cascade failed. The data might be too heterogeneous for a single cascade (existing solutions use separate models for every region's license plate type).
Ready-made openALPR cascade seems to give reasonable results, need to test inference time on Nano
A sit is not possible to run a lightweight face detector alongside, did not test speed on nano
Train and/or use existing cascade detector