2.0.0无法运行模型 - Toybrick

import os
import urllib
import traceback
import time
import sys
import numpy as np
import cv2
import torch
from rknn.api import RKNN
from blazefacenumpy import BlazeFace
ONNX_MODEL = 'best.onnx' # onnx 模型的路径
ONNX_MODEL = 'rppg_new.onnx' # onnx 模型的路径
ONNX_MODEL = 'blazeface128.onnx' # onnx 模型的路径
# ONNX_MODEL = 'v3.onnx' # onnx 模型的路径
# ONNX_MODEL = 'alexnet.onnx' # onnx 模型的路径
TFLITE_MODEL = 'face_detection_front.tflite'
# TFLITE_MODEL = 'alexnet_float32.tflite'
# /anaconda/envs/horizon_bpu/lib/python3.8/site-packages/rknn/3rdparty/platform-tools/adb/linux-x86_64
# tcpip
RKNN_MODEL = './yolov8-ghost-pose.rknn' # 转换后的 RKNN 模型保存路径
RKNN_MODEL = './bz128_3568.rknn' # 转换后的 RKNN 模型保存路径
# RKNN_MODEL = './facenet——3568.rknn' # 转换后的 RKNN 模型保存路径
RKNN_MODEL = './pfldv3.rknn' # 转换后的 RKNN 模型保存路径
DATASET = './test.txt' # 数据集文件路径
QUANTIZE_ON = False # 是否进行量化
def plot_detections(img, detections):
if isinstance(detections, torch.Tensor):
detections = detections.cpu().numpy()
# if detections.ndim == 1:
# detections = np.expand_dims(detections, axis=0)
# print("Found %d faces" % detections.shape[0])
for i in range(detections.shape[0]):
ymin = int(detections[i, 0] * img.shape[0])
xmin = int(detections[i, 1] * img.shape[1])
ymax = int(detections[i, 2] * img.shape[0])
xmax = int(detections[i, 3] * img.shape[1])
cv2.rectangle(img,(xmin,ymin),(xmax,ymax),(170,234,242),5,lineType=cv2.LINE_AA)
# if with_keypoints:
# for k in range(6):
# kp_x = detections[i, 4 + k*2 ] * img.shape[1]
# kp_y = detections[i, 4 + k*2 + 1] * img.shape[0]
# circle = patches.Circle((kp_x, kp_y), radius=0.5, linewidth=1,
# edgecolor="lightskyblue", facecolor="none",
# alpha=detections[i, 16])
# ax.add_patch(circle)
if __name__ == '__main__':
# 创建 RKNN 对象
rknn = RKNN(verbose=False)
# 检查 ONNX 模型文件是否存在
if not os.path.exists(ONNX_MODEL):
print('model not exist')
exit(-1)
# 配置模型预处理参数
print('--> Config model')
rknn.config(#reorder_channel='0 1 2', # 表示 RGB 通道
mean_values=[[0, 0, 0]], # 每个通道的像素均值，预处理时对应通道减去该值
std_values=[[255, 255, 255]], # 每个通道的像素标准差，每个通道除以该值
optimization_level=3, # 优化级别
quantized_method='layer',
float_dtype="float16",
target_platform = 'RK3568', #指定目标平台为rv1126
# quantize_input_node=QUANTIZE_ON
) # 对时输入节点进行量化
# 加载 ONNX 模型
# print('--> Loading model')
model = rknn.load_rknn(RKNN_MODEL)
ret = rknn.init_runtime(target="rk3568")
print("start")
frame = cv2.imread('./test.jpg')
frame =cv2.resize(frame,(112,112))
back_net = BlazeFace(back_model=False).to("cpu")
input = np.reshape(frame,-1)
pred = rknn.inference(inputs=[input], data_format='nhwc')
kk1 = np.reshape( pred[1],-1)[222:225]
print(pred[1].astype(np.float16))
kk =np.reshape( pred[0],-1)[2::10].astype(np.int16)
print('pred')
# print('bz load')
# back_net.load_anchors("anchors.npy")
# deback = back_net._tensors_to_detections(raw_box_tensor=pred[0],raw_score_tensor=pred[1],anchors=back_net.anchors)
# # 4. Non-maximum suppression to remove overlapping detections:
# filtered_detections = []
# for i in range(len(deback)):
# faces = back_net._weighted_non_max_suppression(deback[i])
# faces = np.stack(faces) if len(faces) > 0 else np.zeros((0, 17))
# filtered_detections.append(faces)
# plot_detections(frame,filtered_detections[0])
# cv2.imwrite('test_out.jpg',frame)