Currently Detectree2Detector handles image prediction in call_predict() by passing a list of dictionaries to the model. From my initial understanding, this is a constraint because we are using Detectron2 which expects the input images in this format rather than a Tensor.
However, we want to modify it to take the input as a batch of images and run inference at once so that we can take advantage of the full GPU.
Currently
Detectree2Detectorhandles image prediction incall_predict()by passing a list of dictionaries to the model. From my initial understanding, this is a constraint because we are usingDetectron2which expects the input images in this format rather than aTensor.However, we want to modify it to take the input as a batch of images and run inference at once so that we can take advantage of the full GPU.
(@russelldj)