Mask R-CNN是个好东西,Fine Tuning对数据量要求不多,Inference对资源要求也不高。

标注数据多了之后,模型应该可以自动识别不同车损的类别和程度以及相应的汽车配件。

Automatic Motor Damage Evaluation, using Mask R-CNN for pixel level segmentation, thanks to #tensorflow model zoo.

With more labelled images, the model should be able to automatically identify different types and levels of damage and the corresponding motor parts.