Dogs v Cats

In [1]:
%reload_ext autoreload
%autoreload 2
%matplotlib inline

from fastai.imports import *

from fastai.transforms import *
from fastai.conv_learner import *
from fastai.model import *
from fastai.dataset import *
from fastai.sgdr import *
In [2]:
PATH = "data/dogscats/"
sz = 224
arch = resnet34
bs = 64
In [3]:
m = arch(True)
In [4]:
m
Out[4]:
ResNet(
  (conv1): Conv2d (3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)
  (relu): ReLU(inplace)
  (maxpool): MaxPool2d(kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), dilation=(1, 1))
  (layer1): Sequential(
    (0): BasicBlock(
      (conv1): Conv2d (64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)
    )
    (1): BasicBlock(
      (conv1): Conv2d (64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)
    )
    (2): BasicBlock(
      (conv1): Conv2d (64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True)
    )
  )
  (layer2): Sequential(
    (0): BasicBlock(
      (conv1): Conv2d (64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)
      (downsample): Sequential(
        (0): Conv2d (64, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)
      )
    )
    (1): BasicBlock(
      (conv1): Conv2d (128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)
    )
    (2): BasicBlock(
      (conv1): Conv2d (128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)
    )
    (3): BasicBlock(
      (conv1): Conv2d (128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True)
    )
  )
  (layer3): Sequential(
    (0): BasicBlock(
      (conv1): Conv2d (128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)
      (downsample): Sequential(
        (0): Conv2d (128, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)
      )
    )
    (1): BasicBlock(
      (conv1): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)
    )
    (2): BasicBlock(
      (conv1): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)
    )
    (3): BasicBlock(
      (conv1): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)
    )
    (4): BasicBlock(
      (conv1): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)
    )
    (5): BasicBlock(
      (conv1): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True)
    )
  )
  (layer4): Sequential(
    (0): BasicBlock(
      (conv1): Conv2d (256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)
      (downsample): Sequential(
        (0): Conv2d (256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)
      )
    )
    (1): BasicBlock(
      (conv1): Conv2d (512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)
    )
    (2): BasicBlock(
      (conv1): Conv2d (512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)
      (relu): ReLU(inplace)
      (conv2): Conv2d (512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True)
    )
  )
  (avgpool): AvgPool2d(kernel_size=7, stride=7, padding=0, ceil_mode=False, count_include_pad=True)
  (fc): Linear(in_features=512, out_features=1000)
)
In [5]:
m = nn.Sequential(*children(m)[:-2], 
                  nn.Conv2d(512, 2, 3, padding=1), 
                  nn.AdaptiveAvgPool2d(1), Flatten(), 
                  nn.LogSoftmax())
In [4]:
tfms = tfms_from_model(arch, sz, aug_tfms=transforms_side_on, max_zoom=1.1)
data = ImageClassifierData.from_paths(PATH, tfms=tfms, bs=bs)
In [5]:
learn = ConvLearner.from_model_data(m, data)
In [8]:
learn.freeze_to(-4)
In [9]:
m[-1].trainable
Out[9]:
True
In [10]:
m[-4].trainable
Out[10]:
True
In [11]:
learn.fit(0.01, 1)
[ 0.       0.12421  0.1227   0.98633]                        

In [12]:
learn.fit(0.01, 1, cycle_len=1)
[ 0.       0.0725   0.07043  0.99072]                         

CAM

In [6]:
class SaveFeatures():
    features=None
    def __init__(self, m): self.hook = m.register_forward_hook(self.hook_fn)
    def hook_fn(self, module, input, output): self.features = output
    def remove(self): self.hook.remove()
In [7]:
x,y = next(iter(data.val_dl))
In [10]:
x,y = x[None,1], y[None,1]
vx = Variable(x.cuda(), requires_grad=True)
In [39]:
dx = data.val_ds.denorm(x)[0]
plt.imshow(dx);
In [15]:
sfs = [SaveFeatures(o) for o in [m[-7], m[-6], m[-5], m[-4]]]
In [18]:
%time py = m(Variable(x.cuda()))
Wall time: 20.1 ms
In [19]:
for o in sfs: o.remove()
In [20]:
[o.features.size() for o in sfs]
Out[20]:
[torch.Size([64, 64, 56, 56]),
 torch.Size([64, 128, 28, 28]),
 torch.Size([64, 256, 14, 14]),
 torch.Size([64, 512, 7, 7])]
In [15]:
py = np.exp(to_np(py)[0]); py
Out[15]:
array([ 1.,  0.], dtype=float32)
In [16]:
feat = np.maximum(0,to_np(sfs[3].features[0]))
feat.shape
Out[16]:
(2, 7, 7)
In [23]:
f2=np.dot(np.rollaxis(feat,0,3), py)
f2-=f2.min()
f2/=f2.max()
f2
Out[23]:
array([[ 0.06796,  0.09529,  0.04571,  0.02226,  0.01783,  0.02682,  0.     ],
       [ 0.18074,  0.25366,  0.21349,  0.17134,  0.12896,  0.07998,  0.02269],
       [ 0.32468,  0.53808,  0.58196,  0.51675,  0.36953,  0.20348,  0.06563],
       [ 0.4753 ,  0.82032,  0.93738,  0.84066,  0.58015,  0.30568,  0.09089],
       [ 0.53621,  0.90491,  1.     ,  0.87108,  0.60574,  0.33723,  0.11391],
       [ 0.44609,  0.70038,  0.72927,  0.60194,  0.42627,  0.26258,  0.10247],
       [ 0.21954,  0.3172 ,  0.30497,  0.21987,  0.15255,  0.08741,  0.00709]], dtype=float32)
In [22]:
plt.imshow(dx)
plt.imshow(scipy.misc.imresize(f2, dx.shape), alpha=0.5, cmap='hot');

Model

In [38]:
learn.unfreeze()
learn.bn_freeze(True)
In [39]:
lr=np.array([1e-6,1e-4,1e-2])
In [40]:
learn.fit(lr, 2, cycle_len=1)
[ 0.       0.02049  0.01577  0.99405]                         
[ 1.       0.01648  0.0137   0.99554]                          

In [41]:
accuracy(*learn.TTA())
Out[41]:
0.996
In [42]:
learn.fit(lr, 2, cycle_len=1)
[ 0.       0.01243  0.01535  0.99554]                          
[ 1.       0.0128   0.01507  0.99554]                          

In [43]:
accuracy(*learn.TTA())
Out[43]:
0.99650000000000005
In [ ]: