onnx转caffe前后结果对比

# 1. onnx 推理 demo
import onnxruntime
import numpy as np

session = onnxruntime.InferenceSession(r"./yolov5s-simple-focus.onnx")
img = np.ones((1, 3, 640, 640)).astype(np.float32)

result = session.run(["435"], {"images": img})
print(result[0].shape)
print(result)

# 2. caffe 推理 demo
import sys
CENTERNET_PATH = '/home/zl//yolov5_caffe-master/python'
sys.path.insert(0, CENTERNET_PATH)
import caffe
import numpy as np

weight_file = r"./yolov5s-simple.caffemodel"
model_file = r"./yolov5s-simple.prototxt"
net = caffe.Net(model_file, weight_file, caffe.TEST)
img = np.ones((1, 3, 640, 640))

net.blobs['images'].data[...] = img

out = net.forward()

print(out["432"].shape)
print(out["432"])

网络最后输出的层数可以通过netron查看。