Hi, I have trained a custom yolov5m_vehicles.onnx model and wanted to add in the nms_postprocessing feature so that it can be run on the tappas example. However, I was unable to run the code on DFC successfully, returning me an error of wrong input shape shown as below:
ValueError: Dimension size must be evenly divisible by 115200 but is 2150400 for ‘{{node yolov5m_vehicles/yolov5_nms_postprocess/Reshape}} = Reshape[T=DT_FLOAT, Tshape=DT_INT32](yolov5m_vehicles/yolov5_nms_postprocess/transpose, yolov5m_vehicles/yolov5_nms_postprocess/Reshape/shape)’ with input shapes: [8,42,80,80], [3] and with input tensors computed as partial shapes: input[1] = [?,18,6400].
Call arguments received by layer 'yolov5_nms_postprocess' (type HailoPostprocess):
• inputs=['tf.Tensor(shape=(8, 80, 80, 42), dtype=float32)', 'tf.Tensor(shape=(8, 40, 40, 42), dtype=float32)', 'tf.Tensor(shape=(8, 20, 20, 42), dtype=float32)']
• training=False
• kwargs={'encoding_tensors': 'None'}
Call arguments received by layer 'yolov5m_vehicles' (type HailoModel):
• inputs=tf.Tensor(shape=(8, 640, 640, 3), dtype=float32)
• layer_postproc_cb=None
• training=False
• save_internal_list=None
• ignore_preproc=False
• skip_encoding=False
• kwargs=<class 'inspect._empty'>
May I know how to solve this issue? I want to run the LPR example but I only have Hailo8L accelerator, hence I decided to retrain the model myself to the Hailo8L architecture. Please help, thanks!
