File size: 2,001 Bytes
92f0e98 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 |
'''
Netdissect package.
To run dissection:
1. Load up the convolutional model you wish to dissect, and wrap it
in an InstrumentedModel. Call imodel.retain_layers([layernames,..])
to analyze a specified set of layers.
2. Load the segmentation dataset using the BrodenDataset class;
use the transform_image argument to normalize images to be
suitable for the model, or the size argument to truncate the dataset.
3. Write a function to recover the original image (with RGB scaled to
[0...1]) given a normalized dataset image; ReverseNormalize in this
package inverts transforms.Normalize for this purpose.
4. Choose a directory in which to write the output, and call
dissect(outdir, model, dataset).
Example:
from netdissect import InstrumentedModel, dissect
from netdissect import BrodenDataset, ReverseNormalize
model = InstrumentedModel(load_my_model())
model.eval()
model.cuda()
model.retain_layers(['conv1', 'conv2', 'conv3', 'conv4', 'conv5'])
bds = BrodenDataset('datasets/broden1_227',
transform_image=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(IMAGE_MEAN, IMAGE_STDEV)]),
size=1000)
dissect('result/dissect', model, bds,
recover_image=ReverseNormalize(IMAGE_MEAN, IMAGE_STDEV),
examples_per_unit=10)
'''
__all__ = [
'actviz',
'autoeval',
'bargraph',
'broden',
'customnet',
'easydict',
'encoder_loss',
'encoder_net',
'evalablate',
'frechet_distance',
'fsd',
'fullablate',
'imgsave',
'imgviz',
'invert',
'LBFGS',
'make_z_dataset',
'modelconfig',
'multilayer_graph',
'nethook',
'oldalexnet',
'oldresnet152',
'oldvgg16',
'optimize_residuals',
'optimize_z_lbfgs',
'parallelfolder',
'pbar',
'pidfile',
'plotutil',
'proggan',
'renormalize',
'runningstats',
'samplegan',
'sampler',
'segdata',
'segmenter',
'segviz',
'setting',
'show',
'statedict',
'tally',
'upsample',
'workerpool',
'zdataset',
]
|