def plot_val_with_title(idxs, title):
imgs = np.stack([data.val_ds[x][0] for x in idxs])
title_probs = [probs[x] for x in idxs]
print(title)
return plots(data.val_ds.denorm(imgs), rows=1, titles=title_probs)
def plotXY(x, y):
plots([x])
plt.gca().add_patch(create_rect(y))
def plot_val_with_title(idxs, title):
imgs = [load_img_id(data.val_ds, x) for x in idxs]
title_probs = [probs[x] for x in idxs]
print(title)
return plots(imgs, rows=1, titles=title_probs, figsize=(16, 8))
plotXY(x, y) # ## Scale # In[14]: xx, yy = Scale(sz=350, tfm_y=TfmType.COORD)(x, y) # In[15]: plotXY(xx, yy) # In[16]: xx, yy = Scale(sz=350, tfm_y=TfmType.PIXEL)(x, x) plots([xx, yy]) # ## RandomScale # In[17]: xx, yy = RandomScale(sz=350, max_zoom=1.1, tfm_y=TfmType.COORD)(x, y) plotXY(xx, yy) print(yy) print(y) # In[18]: xx, yy = RandomScale(sz=350, max_zoom=1.1, tfm_y=TfmType.PIXEL)(x, x) plots([xx, yy])
classes=['dogs', 'cats'],
test_name='test',
tfms=tfms,
num_workers=1,
bs=2)
x, _ = next(iter(data.aug_dl))
return data.trn_ds.denorm(x)[1]
# In[ ]:
ims = np.stack([get_augs() for i in range(6)])
# In[ ]:
plots(ims, rows=2)
# Let's create a new `data` object that includes this augmentation in the transforms.
# In[ ]:
data = ImageClassifierData.from_names_and_array(path=PATH,
fnames=fnames,
y=labels,
classes=['dogs', 'cats'],
test_name='test',
tfms=tfms)
learn = ConvLearner.pretrained(arch,
data,
precompute=True,
tmp_name=TMP_PATH,
def plot_aug(tfms):
ims = np.stack([get_augs(tfms) for i in range(6)])
plots(ims, rows=2)
#%%
def plots(ims, figsize=(12, 6), rows=2, titles=None):
f = plt.figure(figsize=figsize)
cols = len(ims) // rows
for i in range(len(ims)):
sp = f.add_subplot(rows, cols, i + 1)
sp.axis('Off')
if titles is not None: sp.set_title(titles[i], fontsize=16)
plt.imshow(ims[i], cmap='gray')
#%%
plots(imgs[:8],
titles=[
class_names[test_preds[0]], class_names[test_preds[1]],
class_names[test_preds[2]], class_names[test_preds[3]],
class_names[test_preds[4]], class_names[test_preds[5]],
class_names[test_preds[6]],
str(data.classes[preds])
])
#%% #Predict on test set This step is under construction...
acc = 0
test_log_preds = np.zeros([7, 60, 7])
preds_test = np.zeros([7, 60, 7])
for insects in data.classes:
data = ImageClassifierData.from_paths(PATH,
bs=bsz,
tfms=tfms_from_model(arch, sz),
test_name='test/' + insects)
learn = ConvLearner.pretrained(arch, data)
test_log_preds[[data.classes.index(insects)]] = learn.predict(is_test=True)
# We can look at part of an image:
# In[19]:
x_imgs[0, 10:15, 10:15]
# In[20]:
show(x_imgs[0, 10:15, 10:15])
# In[21]:
plots(x_imgs[:8],
titles=[
class_names[y_valid[0]], class_names[y_valid[1]],
class_names[y_valid[2]], class_names[y_valid[3]],
class_names[y_valid[4]], class_names[y_valid[5]],
class_names[y_valid[6]], class_names[y_valid[7]]
])
# In[14]:
from fastai.metrics import *
from fastai.model import *
from fastai.dataset import *
from fastai.plots import *
import torch.nn as nn
# We will begin with the highest level abstraction: using a neural net defined by PyTorch's Sequential class.
#%%
