RK3566 Development Board Core Board TC-RK3566 Highlights 1: High performance CPU & GPU TC-RK3566 Highlights 2: New generation (3rd Gen) Rockchip ISP TC-RK3566 Highlight 3: Powerful multimedia decode/encode capability TC-RK3566 Highlight 4: Integrated efficient RKNN AI processing unit
TC-RK3566 Highlights 1: High performance CPU & GPU
New ARM architecture and advanced process brings higher performance and power efficiency
TC-RK3566 Highlights 2: New generation (3rd Gen) Rockchip ISP
up to 8M@30fps processing power, supporting time-sharing and multiplexing dual camera
Powerful HDR function makes the image clear under backlight or strong light conditions
Support dual channel simultaneous zooming output
Noise cancellation function, so that the image under low light conditions is also delicate
Support defogging function, can see clearly even in haze
Support lateral correction of LDCH to remove the distortion caused by the sensor lens
TC-RK3566 Highlight 3: Powerful multimedia decode/encode capability
Support 4KP60 H.264/H.265/VP9 and other formats HD decoding
Support simultaneous decoding of multiple video sources
Support HDR10, excellent performance in color and dynamic range
Support image post-processing, deinterleaving, denoising, color enhancement, resolution increase
Support 1080p 60fps H.264 and H.265 format encoding
Support dynamic bit rate, frame rate, resolution adjustment
TC-RK3566 Highlight 4: Integrated efficient RKNN AI processing unit
NPU with 0.8TOPs computing power
Embedded neural network hardware accelerator, support INT8, INT16, FP16 efficient operation
NPU hardware natively supports technologies such as pre-processing merging, channel quantization, and zero skipping
Support lossless compression of INT8, INT16, FP16 neural network parameters
The NPU core supports ordinary convolution, depth separable convolution, deconvolution, hole convolution, fully connected layer and pooling layer
NPU internal blocks include multiply-add operations, activation, LUT and precision conversion units, and support custom layer construction
Support one-click model conversion, support Caffe/TensorFlow/TF-Lite/ONNX/PyTorch/Keras/Darknet mainstream framework models