Dlr 模型为目标检测模型提供了臭名昭著的结果
Dlr model gives notorious result for object detection model
我正在关注此 link 以训练对象检测模型。我能够在 EC2 实例上成功部署模型。准确性很好。我为 m edge Device Jetson Nano 编写了相同的模型文件。我的推理代码如下所示，
from dlr import DLRModel
import json
import cv2
model = DLRModel('model', 'gpu')
img = cv2.imread('test.jpg')
img = cv2.resize(img, (512, 512), interpolation=cv2.INTER_AREA)
img = np.expand_dims(img, 0)
outputs = model.run(img)
objects=outputs[0][0]
scores=outputs[1][0]
bounding_boxes=outputs[2][0]
当我查看结果时，它与 SageMaker Notebook 实例结果完全不匹配。边界框的值有时在 ~70000 左右。看不懂DLR生成的结果格式。
图像的示例结果。
类:
[[10.0], [14.0], [4.0], [10.0], [14.0], [-1.0], [-1.0], [-1.0], [7.0], [-1.0], [6.0], [ 11.0]、[-1.0]、[-1.0]、[-1.0]、[-1.0]、[-1.0]、[2.0]、[-1.0]、[-1.0]、[0.0]、[-1.0] , [17.0], [-1.0], [-1.0], [6.0], [18.0], [-1.0], [-1.0], [-1.0], [18.0], [-1.0], [12.0] , [-1.0], [-1.0], [13.0], [-1.0], [-1.0], [-1.0], [1.0], [-1.0], [-1.0], [5.0], [- 1.0]、[0.0]、[-1.0]、[-1.0]、[-1.0]、[-1.0]、[3.0]、[8.0]、[5.0]、[-1.0]、[-1.0]、[ 15.0], [-1.0], [9.0], [3.0], [-1.0], [10.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0]、[-1.0]、[16.0]、[8.0]、[-1.0]、[16.0]、[19.0]、[-1.0]、[9.0]、[-1.0]、[4.0]、[ -1.0]、[-1.0]、[15.0]、[10.0]、[-1.0]、[-1.0]、[4.0]、[-1.0]、[8.0]、[2.0]、[-1.0]、[ -1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [ -1.0], [-1.0], [-1.0]]
分数：
[[0.9527158737182617], [0.910746157169342], [0.28013306856155396], [0.059000786393880844], [0.04898739233613014], [-1.0], [-1.0], [-1.0], [0.04864144325256348], [-1.0], [0.04847110062837601], [ 0.04843416064977646]、[-1.0]、[-1.0]、[-1.0]、[-1.0]、[-1.0]、[0.04821664094924927]、[-1.0]、[-1.0]、[0.04808368161320686]、[-1.0] , [0.0479729063808918], [-1.0], [-1.0], [0.04782549664378166], [0.04778601974248886], [-1.0], [-1.0], [-1.0], [0.0475776381790638], [-1.0], [014623475] , [-1.0], [-1.0], [0.047468967735767365], [-1.0], [-1.0], [-1.0], [0.04737424850463867], [-1.0], [-1.0], [0.047330085188150406], [- 1.0], [0.04730956256389618], [-1.0], [-1.0], [-1.0], [-1.0], [0.04710235074162483], [0.04710135608911514], [0.047083333134651184], [-1.0], [-1.0], [ 0.047033149749040604]、[-1.0]、[0.0469636432826519]、[0.046939317137002945]、[-1.0]、[0.04687687009572983]、[-1.0]、[-1.0]、[-1.0]、[-1.0]、[-1.0]、 [-1.0]、[-1.0]、[0.046708278357982635]、[0.046680934727191925]、[-1.0]、[0.0466097407042 9802]、[0.046597886830568314]、[-1.0]、[0.04656397923827171]、[-1.0]、[0.046513814479112625]、[-1.0]、[-1.0]、[0.04647510126233101]、[0.0464469]、[0.04644694] 1.0], [0.046245038509368896], [-1.0], [0.01647786796092987], [0.010514501482248306], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0]]
布定框：
[[523.2319946289062, -777751.875, 523.5722045898438, 778992.625], [425.0546875, -1141093.0, 429.6099853515625, 1142524.5], [680.2073364257812, 542.9566650390625, 680.2202758789062, 543.36376953125], [425.0546875, -1141093.0, 429.6099853515625, 1142524.5], [21.017044067382812, 20.89019775390625, 491.50994873046875 , 491.19842529296875], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [30.993896484375, 30.6316528320 3125, 480.84320068359375, 481.7999267578125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [ -60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875] , [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.73840332 03125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [-60.170318603515625, 95.33354187011719 , 571.584716796875, 417.34515380859375], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 4 80.84320068359375, 481.7999267578125], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [- 60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125] , [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [-60.17031 8603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [21.017044067382812, 20.89019775390625, 491.50994873046 875, 491.19842529296875], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [ -60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [ 30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [591.7652587890625, -1121890.0, 592.9493408203125, 1123299.75], [523.2319946289062, -777751.875, 523.5722045898438, 778992.625]、[-1.0、-1.0、-1.0、-1.0]、[-1.0、-1.0、-1.0、-1.0]、[-1.0、-1.0、-1.0、-1.0]、[-1.0、- 1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1.0] , [-1.0, -1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1 .0], [-1.0, -1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1.0]]
是什么导致了这个臭名昭著的结果？
在 Neo 中编译模型时有什么问题吗？
推理代码有什么问题吗？
任何提示都将不胜感激。
结果异常的原因是预处理方法不当。这是Mobilenet-ssd模型的完整推理代码。
def transform(img):
# Normalize
mean_vec = np.array([0.485, 0.456, 0.406])
stddev_vec = np.array([0.229, 0.224, 0.225])
image = (img / 255 - mean_vec) / stddev_vec
# Transpose
if len(image.shape) == 2: # for greyscale image
image = np.expand_dims(image, axis=2)
image = np.rollaxis(image, axis=2, start=0)[np.newaxis, :]
return image
model = DLRModel(model_dir, 'gpu')
for file_name in image_folder:
image = PIL.Image.open(file_name)
image = np.asarray(image.resize((512, 512)))
image = transform(image)
# flatten within a input array
input_data = {'data': image}
outputs = model.run(input_data)
objects = outputs[0]
scores = outputs[1]
bounding_boxes = outputs[2]
result = [objects.tolist(), scores.tolist(), bounding_boxes.tolist()]
print(json.dumps(result))
我正在关注此 link 以训练对象检测模型。我能够在 EC2 实例上成功部署模型。准确性很好。我为 m edge Device Jetson Nano 编写了相同的模型文件。我的推理代码如下所示，
from dlr import DLRModel
import json
import cv2
model = DLRModel('model', 'gpu')
img = cv2.imread('test.jpg')
img = cv2.resize(img, (512, 512), interpolation=cv2.INTER_AREA)
img = np.expand_dims(img, 0)
outputs = model.run(img)
objects=outputs[0][0]
scores=outputs[1][0]
bounding_boxes=outputs[2][0]
当我查看结果时，它与 SageMaker Notebook 实例结果完全不匹配。边界框的值有时在 ~70000 左右。看不懂DLR生成的结果格式。
图像的示例结果。
类: [[10.0], [14.0], [4.0], [10.0], [14.0], [-1.0], [-1.0], [-1.0], [7.0], [-1.0], [6.0], [ 11.0]、[-1.0]、[-1.0]、[-1.0]、[-1.0]、[-1.0]、[2.0]、[-1.0]、[-1.0]、[0.0]、[-1.0] , [17.0], [-1.0], [-1.0], [6.0], [18.0], [-1.0], [-1.0], [-1.0], [18.0], [-1.0], [12.0] , [-1.0], [-1.0], [13.0], [-1.0], [-1.0], [-1.0], [1.0], [-1.0], [-1.0], [5.0], [- 1.0]、[0.0]、[-1.0]、[-1.0]、[-1.0]、[-1.0]、[3.0]、[8.0]、[5.0]、[-1.0]、[-1.0]、[ 15.0], [-1.0], [9.0], [3.0], [-1.0], [10.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0]、[-1.0]、[16.0]、[8.0]、[-1.0]、[16.0]、[19.0]、[-1.0]、[9.0]、[-1.0]、[4.0]、[ -1.0]、[-1.0]、[15.0]、[10.0]、[-1.0]、[-1.0]、[4.0]、[-1.0]、[8.0]、[2.0]、[-1.0]、[ -1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [ -1.0], [-1.0], [-1.0]]
分数： [[0.9527158737182617], [0.910746157169342], [0.28013306856155396], [0.059000786393880844], [0.04898739233613014], [-1.0], [-1.0], [-1.0], [0.04864144325256348], [-1.0], [0.04847110062837601], [ 0.04843416064977646]、[-1.0]、[-1.0]、[-1.0]、[-1.0]、[-1.0]、[0.04821664094924927]、[-1.0]、[-1.0]、[0.04808368161320686]、[-1.0] , [0.0479729063808918], [-1.0], [-1.0], [0.04782549664378166], [0.04778601974248886], [-1.0], [-1.0], [-1.0], [0.0475776381790638], [-1.0], [014623475] , [-1.0], [-1.0], [0.047468967735767365], [-1.0], [-1.0], [-1.0], [0.04737424850463867], [-1.0], [-1.0], [0.047330085188150406], [- 1.0], [0.04730956256389618], [-1.0], [-1.0], [-1.0], [-1.0], [0.04710235074162483], [0.04710135608911514], [0.047083333134651184], [-1.0], [-1.0], [ 0.047033149749040604]、[-1.0]、[0.0469636432826519]、[0.046939317137002945]、[-1.0]、[0.04687687009572983]、[-1.0]、[-1.0]、[-1.0]、[-1.0]、[-1.0]、 [-1.0]、[-1.0]、[0.046708278357982635]、[0.046680934727191925]、[-1.0]、[0.0466097407042 9802]、[0.046597886830568314]、[-1.0]、[0.04656397923827171]、[-1.0]、[0.046513814479112625]、[-1.0]、[-1.0]、[0.04647510126233101]、[0.0464469]、[0.04644694] 1.0], [0.046245038509368896], [-1.0], [0.01647786796092987], [0.010514501482248306], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0], [-1.0]]
布定框： [[523.2319946289062, -777751.875, 523.5722045898438, 778992.625], [425.0546875, -1141093.0, 429.6099853515625, 1142524.5], [680.2073364257812, 542.9566650390625, 680.2202758789062, 543.36376953125], [425.0546875, -1141093.0, 429.6099853515625, 1142524.5], [21.017044067382812, 20.89019775390625, 491.50994873046875 , 491.19842529296875], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [30.993896484375, 30.6316528320 3125, 480.84320068359375, 481.7999267578125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [ -60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875] , [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.73840332 03125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [-60.170318603515625, 95.33354187011719 , 571.584716796875, 417.34515380859375], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 4 80.84320068359375, 481.7999267578125], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [- 60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125] , [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [-60.17031 8603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [21.017044067382812, 20.89019775390625, 491.50994873046 875, 491.19842529296875], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [ -60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [94.88882446289062, -59.820770263671875, 417.0069885253906, 571.7384033203125], [21.017044067382812, 20.89019775390625, 491.50994873046875, 491.19842529296875], [ 30.993896484375, 30.63165283203125, 480.84320068359375, 481.7999267578125], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [-60.170318603515625, 95.33354187011719, 571.584716796875, 417.34515380859375], [591.7652587890625, -1121890.0, 592.9493408203125, 1123299.75], [523.2319946289062, -777751.875, 523.5722045898438, 778992.625]、[-1.0、-1.0、-1.0、-1.0]、[-1.0、-1.0、-1.0、-1.0]、[-1.0、-1.0、-1.0、-1.0]、[-1.0、- 1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1.0] , [-1.0, -1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1 .0], [-1.0, -1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1.0], [-1.0, -1.0, -1.0, -1.0]]
是什么导致了这个臭名昭著的结果？ 在 Neo 中编译模型时有什么问题吗？ 推理代码有什么问题吗？
任何提示都将不胜感激。
结果异常的原因是预处理方法不当。这是Mobilenet-ssd模型的完整推理代码。
def transform(img):
# Normalize
mean_vec = np.array([0.485, 0.456, 0.406])
stddev_vec = np.array([0.229, 0.224, 0.225])
image = (img / 255 - mean_vec) / stddev_vec
# Transpose
if len(image.shape) == 2: # for greyscale image
image = np.expand_dims(image, axis=2)
image = np.rollaxis(image, axis=2, start=0)[np.newaxis, :]
return image
model = DLRModel(model_dir, 'gpu')
for file_name in image_folder:
image = PIL.Image.open(file_name)
image = np.asarray(image.resize((512, 512)))
image = transform(image)
# flatten within a input array
input_data = {'data': image}
outputs = model.run(input_data)
objects = outputs[0]
scores = outputs[1]
bounding_boxes = outputs[2]
result = [objects.tolist(), scores.tolist(), bounding_boxes.tolist()]
print(json.dumps(result))