Object detection system and object detection method

An object detection system, comprising:

• an image capture device, configured to capture an image;
• a memory, configured to store at least one instruction corresponding to an inference engine, wherein the inference engine is based on a convolutional neural network architecture having multiple scales, the multiple scales of the convolutional neural network architecture comprises a first scale receiving an initial input, a second scale connected to the first scale, and an object detection scale connected to the second scale, a plurality of first convolution layers of the first scale corresponds to a first network width, and a plurality of second convolution layers of the second scale corresponds to a second network width;
• and
• a processor, coupled to the memory and the image capture device, wherein the processor is configured to access the at least one instruction from the memory and execute the at least one instruction to:
• adjust the convolutional neural network architecture to reduce the second network width of the plurality of second convolution layers;
• run the inference engine according to the adjusted convolutional neural network architecture, to enable the first scale to receive the initial input, wherein the initial input is information of the image;
• generate, a first output according to the initial input by the first scale, and input the first output into the second scale and the object detection scale;
• generate a second output according to the first output by the second scale, and input the second output into the object detection scale;
• and
• generate a final output according to the first output and the second output by the object detection scale, to perform object detection on the image.