同步操作将从 NVIDIA_Developer_Community/DIGITS 强制同步,此操作会覆盖自 Fork 仓库以来所做的任何修改,且无法恢复!!!
确定后同步将在后台操作,完成时将刷新页面,请耐心等待。
This example shows how to use python to consume a DIGITS model and classify images. You should see the exact same numerical answers between DIGITS and these scripts (but take note of the "Limitations" section).
There are a few "gotchas" which can lead to discrepancies between DIGITS classifications and the output from this example.
This example resizes images using the Squash
method.
In DIGITS, the same method will be used as was originally used when creating your dataset (Crop
, Squash
, Fill
or HalfCrop
).
This example subtracts a mean pixel rather than the whole mean file.
In DIGITS, the same method will be used as was originally used when training your model (None
, Image
or Pixel
).
See BuildCaffe.md for instructions about installing caffe.
Other requirements can be found in requirements.txt
.
You do not need as many packages to run this example as you do to run DIGITS.
Use one of the two python scripts provided to classify an image.
Use use_archive.py
to classify images using a model archive downloaded from DIGITS (e.g. 20150512-171624-d9a9_epoch_30.tar.gz
).
$ ./use_archive.py -h
usage: use_archive.py [-h] [--nogpu] archive image
Classification example using an archive - DIGITS
positional arguments:
archive Path to a DIGITS model archive
image Path to an image
optional arguments:
-h, --help show this help message and exit
--nogpu Don't use the GPU
$ ./use_archive.py digits-model.tar.gz test-image.jpg
Extracting tarfile ...
Processed 1/1 images ...
Classification took 0.00310683250427 seconds.
--------------------------- Prediction for image.jpg ---------------------------
96.1199% - "0"
1.3588% - "6"
0.7247% - "9"
0.4695% - "2"
0.3857% - "3"
Script took 0.270452022552 seconds.
If you have already extracted your model, you can specify each of the files manually with example.py
.
$ ./example.py -h
usage: example.py [-h] [-m MEAN] [-l LABELS] [--nogpu]
caffemodel deploy_file image
Classification example - DIGITS
positional arguments:
caffemodel Path to a .caffemodel
deploy_file Path to the deploy file
image Path to an image
optional arguments:
-h, --help show this help message and exit
-m MEAN, --mean MEAN Path to a mean file (*.npy)
-l LABELS, --labels LABELS
Path to a labels file
--nogpu Don't use the GPU
$ ./example.py snapshot_iter_1000.caffemodel deploy.prototxt test-image.jpg --mean mean.binaryproto --labels labels.txt
Processed 1/1 images ...
Classification took 0.00309991836548 seconds.
--------------------------- Prediction for image.jpg ---------------------------
96.1199% - "0"
1.3588% - "6"
0.7247% - "9"
0.4695% - "2"
0.3857% - "3"
Script took 0.269672870636 seconds.
This example is kept pretty basic. The user is encouraged to extend it to suit their own purposes.
The code already supports classifying a list of images, you just have to provide your own code to give the list of filenames.
Already done, just provide a value other than 1 for batch_size
in classify()
.
See Caffe's example for extracting filters from a trained model: https://github.com/BVLC/caffe/blob/rc2/examples/filter_visualization.ipynb
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。