def viz_cam(model_file, data_dir):
ds = get_data('val')
pred_config = PredictConfig(
model=Model(),
session_init=SmartInit(model_file),
input_names=['input', 'label'],
output_names=['wrong-top1', 'group3new/bnlast/Relu', 'linearnew/W'],
return_input=True)
meta = dataset.ILSVRCMeta().get_synset_words_1000()
pred = SimpleDatasetPredictor(pred_config, ds)
cnt = 0
for inp, outp in pred.get_result():
images, labels = inp
wrongs, convmaps, W = outp
batch = wrongs.shape[0]
for i in range(batch):
if wrongs[i]:
continue
weight = W[:, [labels[i]]].T # 512x1
convmap = convmaps[i, :, :, :] # 512xhxw
mergedmap = np.matmul(weight, convmap.reshape(
(512, -1))).reshape(14, 14)
mergedmap = cv2.resize(mergedmap, (224, 224))
heatmap = viz.intensity_to_rgb(mergedmap, normalize=True)
blend = images[i] * 0.5 + heatmap * 0.5
concat = np.concatenate((images[i], heatmap, blend), axis=1)
classname = meta[labels[i]].split(',')[0]
cv2.imwrite('cam{}-{}.jpg'.format(cnt, classname), concat)
cnt += 1
if cnt == 500:
return
def get_data(train_or_test):
isTrain = train_or_test == 'train'
ds = dataset.ILSVRC12(args.data,
train_or_test,
shuffle=True if isTrain else False)
meta = dataset.ILSVRCMeta()
pp_mean = meta.get_per_pixel_mean()
if isTrain:
# TODO use the augmentor in GoogleNet
augmentors = [
imgaug.Resize((256, 256)),
imgaug.Brightness(30, False),
imgaug.Contrast((0.8, 1.2), True),
imgaug.MapImage(lambda x: x - pp_mean),
imgaug.RandomCrop((224, 224)),
imgaug.Flip(horiz=True),
]
else:
augmentors = [
imgaug.Resize((256, 256)),
imgaug.MapImage(lambda x: x - pp_mean),
imgaug.CenterCrop((224, 224)),
]
ds = AugmentImageComponent(ds, augmentors, copy=False)
ds = BatchData(ds, BATCH_SIZE, remainder=not isTrain)
if isTrain:
ds = PrefetchDataZMQ(ds, 6)
return ds
def get_imglist(dir, name):
"""
Args:
dir(str): directory which contains name
name(str): 'train' or 'val'
Returns:
[(full filename, label)]
"""
dir = os.path.join(dir, name)
meta = dataset.ILSVRCMeta()
imglist = meta.get_image_list(name,
dataset.ILSVRCMeta.guess_dir_structure(dir))
def _filter(fname):
# png
return 'n02105855_2933.JPEG' in fname
ret = []
for fname, label in imglist:
if _filter(fname):
logger.info("Image {} was filtered out.".format(fname))
continue
fname = os.path.join(dir, fname)
ret.append((fname, label))
return ret
def run_image(model, sess_init, inputs):
pred_config = PredictConfig(
model=model,
session_init=sess_init,
input_names=['input'],
output_names=['output']
)
predictor = OfflinePredictor(pred_config)
meta = dataset.ILSVRCMeta()
words = meta.get_synset_words_1000()
transformers = imgaug.AugmentorList(fbresnet_augmentor(isTrain=False))
for f in inputs:
assert os.path.isfile(f), f
img = cv2.imread(f).astype('float32')
assert img is not None
img = transformers.augment(img)[np.newaxis, :, :, :]
outputs = predictor(img)[0]
prob = outputs[0]
ret = prob.argsort()[-10:][::-1]
names = [words[i] for i in ret]
print(f + ":")
print(list(zip(names, prob[ret])))
def get_data(train_or_test):
isTrain = train_or_test == 'train'
ds = dataset.ILSVRC12(args.data, train_or_test,
shuffle=True if isTrain else False)
meta = dataset.ILSVRCMeta()
pp_mean = meta.get_per_pixel_mean()
pp_mean_299 = cv2.resize(pp_mean, (299, 299))
if isTrain:
class Resize(imgaug.ImageAugmentor):
def __init__(self):
self._init(locals())
def _augment(self, img, _):
h, w = img.shape[:2]
size = 299
scale = self.rng.randint(size, 340) * 1.0 / min(h, w)
scaleX = scale * self.rng.uniform(0.85, 1.15)
scaleY = scale * self.rng.uniform(0.85, 1.15)
desSize = map(int, (max(size, min(w, scaleX * w)),
max(size, min(h, scaleY * h))))
dst = cv2.resize(img, tuple(desSize), interpolation=cv2.INTER_CUBIC)
return dst
augmentors = [
Resize(),
imgaug.Rotation(max_deg=10),
imgaug.RandomApplyAug(imgaug.GaussianBlur(3), 0.5),
imgaug.Brightness(30, True),
imgaug.Gamma(),
imgaug.Contrast((0.8, 1.2), True),
imgaug.RandomCrop((299, 299)),
imgaug.RandomApplyAug(imgaug.JpegNoise(), 0.8),
imgaug.RandomApplyAug(imgaug.GaussianDeform(
[(0.2, 0.2), (0.2, 0.8), (0.8, 0.8), (0.8, 0.2)],
(299, 299), 0.2, 3), 0.1),
imgaug.Flip(horiz=True),
imgaug.MapImage(lambda x: x - pp_mean_299),
]
else:
def resize_func(im):
h, w = im.shape[:2]
scale = 340.0 / min(h, w)
desSize = map(int, (max(299, min(w, scale * w)),
max(299, min(h, scale * h))))
im = cv2.resize(im, tuple(desSize), interpolation=cv2.INTER_CUBIC)
return im
augmentors = [
imgaug.MapImage(resize_func),
imgaug.CenterCrop((299, 299)),
imgaug.MapImage(lambda x: x - pp_mean_299),
]
ds = AugmentImageComponent(ds, augmentors, copy=False)
ds = BatchData(ds, BATCH_SIZE, remainder=not isTrain)
if isTrain:
ds = PrefetchDataZMQ(ds, min(12, multiprocessing.cpu_count()))
return ds
def get_data(train_or_test):
isTrain = train_or_test == 'train'
augs = fbresnet_augmentor(isTrain)
meta = dataset.ILSVRCMeta()
pp_mean = meta.get_per_pixel_mean()
augs.append(imgaug.MapImage(lambda x: x - pp_mean[16:-16, 16:-16]))
ds = get_imagenet_dataflow(args.data, train_or_test, BATCH_SIZE, augs)
return ds
def get_data(train_or_test):
isTrain = train_or_test == 'train'
augs = fbresnet_augmentor(isTrain)
meta = dataset.ILSVRCMeta()
pp_mean = meta.get_per_pixel_mean()
augs.append(imgaug.MapImage(lambda x: x - pp_mean[16:-16, 16:-16]))
data_path = TRAIN_LIST_PATH if train_or_test == 'train' else VALID_LIST_PATH
ds = get_imagenet_dataflow(data_path, train_or_test, BATCH_SIZE, augs)
return ds
def run_image(model, sess_init, inputs):
pred_config = PredictConfig(
model=model,
session_init=sess_init,
input_names=['input'],
output_names=['output']
)
predictor = OfflinePredictor(pred_config)
meta = dataset.ILSVRCMeta()
pp_mean = meta.get_per_pixel_mean()
pp_mean_224 = pp_mean[16:-16, 16:-16, :]
words = meta.get_synset_words_1000()
def resize_func(im):
h, w = im.shape[:2]
scale = 256.0 / min(h, w)
desSize = map(int, (max(224, min(w, scale * w)),
max(224, min(h, scale * h))))
im = cv2.resize(im, tuple(desSize), interpolation=cv2.INTER_CUBIC)
return im
transformers = imgaug.AugmentorList([
imgaug.MapImage(resize_func),
imgaug.CenterCrop((224, 224)),
imgaug.MapImage(lambda x: x - pp_mean_224),
])
for f in inputs:
assert os.path.isfile(f)
img = cv2.imread(f).astype('float32')
assert img is not None
img = transformers.augment(img)[np.newaxis, :, :, :]
outputs = predictor(img)[0]
prob = outputs[0]
ret = prob.argsort()[-10:][::-1]
names = [words[i] for i in ret]
print(f + ":")
print(list(zip(names, prob[ret])))
