We hope to achieve an chest Xray detecting system
The top-1/5 accuracy rates by using single center crop (crop size: 224x224, image size: 256xN)
Network | Top-1 | Top-5 | Theano | Tensorflow |
---|---|---|---|---|
DenseNet 121 (k=32) | 74.91 | 92.19 | model (32 MB) | model (32 MB) |
DenseNet 169 (k=32) | 76.09 | 93.14 | model (56 MB) | model (56 MB) |
DenseNet 161 (k=48) | 77.64 | 93.79 | model (112 MB) | model (112 MB) |
First, download the above pretrained weights to the imagenet_models
folder.
Run test_inference.py
for an example of how to use the pretrained model to make inference.
python test_inference.py
for image viewer, write code:
d1 = dict[b'data'][1] print(d1) NewImage = Image.new('RGB', (32, 32)) print(d1[0]) pixelList = []
for i in range(0, 1024):
pixel = (d1[i], d1[i + 1024], d1[i + 2048])
pixelList.append(pixel)
print(pixelList)
NewImage.putdata(pixelList)
NewImage.save('test5.jpg')
Feel free to modify the code to your need