Files
rknn-models/conv.py
2023-11-15 00:00:21 +01:00

61 lines
1.7 KiB
Python

from rknn.api import RKNN
INPUT_MODEL = "yolov8x.onnx"
WIDTH = 320
HEIGHT = 320
OUTPUT_MODEL_BASENAME = 'yolov8x'
QUANTIZATION = False
DATASET = './dataset_coco10.txt'
# Config
MEAN_VALUES = [[0, 0, 0]]
STD_VALUES = [[255, 255, 255]]
QUANT_IMG_RGB2BGR = True
QUANTIZED_DTYPE = "asymmetric_quantized-8"
QUANTIZED_ALGORITHM = "normal"
QUANTIZED_METHOD = "channel"
FLOAT_DTYPE = "float16"
OPTIMIZATION_LEVEL = 2
TARGET_PLATFORM = "rk3588"
CUSTOM_STRING = None
REMOVE_WEIGHT = None
COMPRESS_WEIGHT = False
SINGLE_CORE_MODE = False
MODEL_PRUNNING = False
OP_TARGET = None
DYNAMIC_INPUT = None
OUTPUT_MODEL = OUTPUT_MODEL_BASENAME + '-' + str(WIDTH) + 'x' + str(HEIGHT) + ".rknn"
rknn = RKNN()
rknn.config(mean_values=MEAN_VALUES,
std_values=STD_VALUES,
quant_img_RGB2BGR=QUANT_IMG_RGB2BGR,
quantized_dtype=QUANTIZED_DTYPE,
quantized_algorithm=QUANTIZED_ALGORITHM,
quantized_method=QUANTIZED_METHOD,
float_dtype=FLOAT_DTYPE,
optimization_level=OPTIMIZATION_LEVEL,
target_platform=TARGET_PLATFORM,
custom_string=CUSTOM_STRING,
remove_weight=REMOVE_WEIGHT,
compress_weight=COMPRESS_WEIGHT,
single_core_mode=SINGLE_CORE_MODE,
model_pruning=MODEL_PRUNNING,
op_target=OP_TARGET,
dynamic_input=DYNAMIC_INPUT)
# if rknn.load_pytorch("./input/" + INPUT_MODEL, [[HEIGHT, WIDTH, 3]]) != 0:
if rknn.load_onnx("./input/" + INPUT_MODEL) != 0:
print('Error loading model.')
exit()
if rknn.build(do_quantization=QUANTIZATION, dataset=DATASET) != 0:
print('Error building model.')
exit()
if rknn.export_rknn("./output/" + OUTPUT_MODEL) != 0:
print('Error exporting rknn model.')
exit()