Bienqq
/
car-detection-bayes
2
1
Fork 0
Code Issues Pull Requests Releases Wiki Activity

uncached label removal

This commit is contained in:
Glenn Jocher 2020-04-28 11:07:26 -07:00
parent b1d385a8de
commit 15f1343dfc
1 changed files with 75 additions and 86 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

161
utils/datasets.py
View File

@ -257,7 +257,7 @@ class LoadStreams: # multiple IP or RTSP cameras
class LoadImagesAndLabels(Dataset): # for training/testing class LoadImagesAndLabels(Dataset): # for training/testing
def __init__(self, path, img_size=416, batch_size=16, augment=False, hyp=None, rect=False, image_weights=False, def __init__(self, path, img_size=416, batch_size=16, augment=False, hyp=None, rect=False, image_weights=False,
cache_labels=True, cache_images=False, single_cls=False): cache_images=False, single_cls=False):
path = str(Path(path)) # os-agnostic path = str(Path(path)) # os-agnostic
assert os.path.isfile(path), 'File not found %s. See %s' % (path, help_url) assert os.path.isfile(path), 'File not found %s. See %s' % (path, help_url)
with open(path, 'r') as f: with open(path, 'r') as f:
@ -315,71 +315,69 @@ class LoadImagesAndLabels(Dataset): # for training/testing
self.batch_shapes = np.ceil(np.array(shapes) * img_size / 64.).astype(np.int) * 64 self.batch_shapes = np.ceil(np.array(shapes) * img_size / 64.).astype(np.int) * 64
# Preload labels (required for weighted CE training) # Cache labels
self.imgs = [None] * n self.imgs = [None] * n
self.labels = [None] * n self.labels = [np.zeros((0, 5), dtype=np.float32)] * n
if cache_labels or image_weights: # cache labels for faster training extract_bounding_boxes = False
self.labels = [np.zeros((0, 5))] * n create_datasubset = False
extract_bounding_boxes = False pbar = tqdm(self.label_files, desc='Caching labels')
create_datasubset = False nm, nf, ne, ns, nd = 0, 0, 0, 0, 0 # number missing, found, empty, datasubset, duplicate
pbar = tqdm(self.label_files, desc='Caching labels') for i, file in enumerate(pbar):
nm, nf, ne, ns, nd = 0, 0, 0, 0, 0 # number missing, found, empty, datasubset, duplicate try:
for i, file in enumerate(pbar): with open(file, 'r') as f:
try: l = np.array([x.split() for x in f.read().splitlines()], dtype=np.float32)
with open(file, 'r') as f: except:
l = np.array([x.split() for x in f.read().splitlines()], dtype=np.float32) nm += 1 # print('missing labels for image %s' % self.img_files[i]) # file missing
except: continue
nm += 1 # print('missing labels for image %s' % self.img_files[i]) # file missing
continue
if l.shape[0]: if l.shape[0]:
assert l.shape[1] == 5, '> 5 label columns: %s' % file assert l.shape[1] == 5, '> 5 label columns: %s' % file
assert (l >= 0).all(), 'negative labels: %s' % file assert (l >= 0).all(), 'negative labels: %s' % file
assert (l[:, 1:] <= 1).all(), 'non-normalized or out of bounds coordinate labels: %s' % file assert (l[:, 1:] <= 1).all(), 'non-normalized or out of bounds coordinate labels: %s' % file
if np.unique(l, axis=0).shape[0] < l.shape[0]: # duplicate rows if np.unique(l, axis=0).shape[0] < l.shape[0]: # duplicate rows
nd += 1 # print('WARNING: duplicate rows in %s' % self.label_files[i]) # duplicate rows nd += 1 # print('WARNING: duplicate rows in %s' % self.label_files[i]) # duplicate rows
if single_cls: if single_cls:
l[:, 0] = 0 # force dataset into single-class mode l[:, 0] = 0 # force dataset into single-class mode
self.labels[i] = l self.labels[i] = l
nf += 1 # file found nf += 1 # file found
# Create subdataset (a smaller dataset) # Create subdataset (a smaller dataset)
if create_datasubset and ns < 1E4: if create_datasubset and ns < 1E4:
if ns == 0: if ns == 0:
create_folder(path='./datasubset') create_folder(path='./datasubset')
os.makedirs('./datasubset/images') os.makedirs('./datasubset/images')
exclude_classes = 43 exclude_classes = 43
if exclude_classes not in l[:, 0]: if exclude_classes not in l[:, 0]:
ns += 1 ns += 1
# shutil.copy(src=self.img_files[i], dst='./datasubset/images/') # copy image # shutil.copy(src=self.img_files[i], dst='./datasubset/images/') # copy image
with open('./datasubset/images.txt', 'a') as f: with open('./datasubset/images.txt', 'a') as f:
f.write(self.img_files[i] + '\n') f.write(self.img_files[i] + '\n')
# Extract object detection boxes for a second stage classifier # Extract object detection boxes for a second stage classifier
if extract_bounding_boxes: if extract_bounding_boxes:
p = Path(self.img_files[i]) p = Path(self.img_files[i])
img = cv2.imread(str(p)) img = cv2.imread(str(p))
h, w = img.shape[:2] h, w = img.shape[:2]
for j, x in enumerate(l): for j, x in enumerate(l):
f = '%s%sclassifier%s%g_%g_%s' % (p.parent.parent, os.sep, os.sep, x[0], j, p.name) f = '%s%sclassifier%s%g_%g_%s' % (p.parent.parent, os.sep, os.sep, x[0], j, p.name)
if not os.path.exists(Path(f).parent): if not os.path.exists(Path(f).parent):
os.makedirs(Path(f).parent) # make new output folder os.makedirs(Path(f).parent) # make new output folder
b = x[1:] * [w, h, w, h] # box b = x[1:] * [w, h, w, h] # box
b[2:] = b[2:].max() # rectangle to square b[2:] = b[2:].max() # rectangle to square
b[2:] = b[2:] * 1.3 + 30 # pad b[2:] = b[2:] * 1.3 + 30 # pad
b = xywh2xyxy(b.reshape(-1, 4)).ravel().astype(np.int) b = xywh2xyxy(b.reshape(-1, 4)).ravel().astype(np.int)
b[[0, 2]] = np.clip(b[[0, 2]], 0, w) # clip boxes outside of image b[[0, 2]] = np.clip(b[[0, 2]], 0, w) # clip boxes outside of image
b[[1, 3]] = np.clip(b[[1, 3]], 0, h) b[[1, 3]] = np.clip(b[[1, 3]], 0, h)
assert cv2.imwrite(f, img[b[1]:b[3], b[0]:b[2]]), 'Failure extracting classifier boxes' assert cv2.imwrite(f, img[b[1]:b[3], b[0]:b[2]]), 'Failure extracting classifier boxes'
else: else:
ne += 1 # print('empty labels for image %s' % self.img_files[i]) # file empty ne += 1 # print('empty labels for image %s' % self.img_files[i]) # file empty
# os.system("rm '%s' '%s'" % (self.img_files[i], self.label_files[i])) # remove # os.system("rm '%s' '%s'" % (self.img_files[i], self.label_files[i])) # remove
pbar.desc = 'Caching labels (%g found, %g missing, %g empty, %g duplicate, for %g images)' % ( pbar.desc = 'Caching labels (%g found, %g missing, %g empty, %g duplicate, for %g images)' % (
nf, nm, ne, nd, n) nf, nm, ne, nd, n)
assert nf > 0, 'No labels found in %s. See %s' % (os.path.dirname(file) + os.sep, help_url) assert nf > 0, 'No labels found in %s. See %s' % (os.path.dirname(file) + os.sep, help_url)
# Cache images into memory for faster training (WARNING: large datasets may exceed system RAM) # Cache images into memory for faster training (WARNING: large datasets may exceed system RAM)
if cache_images: # if training if cache_images: # if training
@ -432,7 +430,7 @@ class LoadImagesAndLabels(Dataset): # for training/testing
# Load labels # Load labels
labels = [] labels = []
x = self.labels[index] x = self.labels[index]
if x is not None and x.size > 0: if x.size > 0:
# Normalized xywh to pixel xyxy format # Normalized xywh to pixel xyxy format
labels = x.copy() labels = x.copy()
labels[:, 1] = ratio[0] * w * (x[:, 1] - x[:, 3] / 2) + pad[0] # pad width labels[:, 1] = ratio[0] * w * (x[:, 1] - x[:, 3] / 2) + pad[0] # pad width
@ -502,9 +500,9 @@ def load_image(self, index):
# loads 1 image from dataset, returns img, original hw, resized hw # loads 1 image from dataset, returns img, original hw, resized hw
img = self.imgs[index] img = self.imgs[index]
if img is None: # not cached if img is None: # not cached
img_path = self.img_files[index] path = self.img_files[index]
img = cv2.imread(img_path) # BGR img = cv2.imread(path) # BGR
assert img is not None, 'Image Not Found ' + img_path assert img is not None, 'Image Not Found ' + path
h0, w0 = img.shape[:2] # orig hw h0, w0 = img.shape[:2] # orig hw
r = self.img_size / max(h0, w0) # resize image to img_size r = self.img_size / max(h0, w0) # resize image to img_size
if r < 1 or (self.augment and r != 1): # always resize down, only resize up if training with augmentation if r < 1 or (self.augment and r != 1): # always resize down, only resize up if training with augmentation
@ -557,24 +555,15 @@ def load_mosaic(self, index):
padw = x1a - x1b padw = x1a - x1b
padh = y1a - y1b padh = y1a - y1b
# Load labels # Labels
label_path = self.label_files[index] x = self.labels[index]
if os.path.isfile(label_path): labels = x.copy()
x = self.labels[index] if x.size > 0: # Normalized xywh to pixel xyxy format
if x is None: # labels not preloaded labels[:, 1] = w * (x[:, 1] - x[:, 3] / 2) + padw
with open(label_path, 'r') as f: labels[:, 2] = h * (x[:, 2] - x[:, 4] / 2) + padh
x = np.array([x.split() for x in f.read().splitlines()], dtype=np.float32) labels[:, 3] = w * (x[:, 1] + x[:, 3] / 2) + padw
labels[:, 4] = h * (x[:, 2] + x[:, 4] / 2) + padh
if x.size > 0: labels4.append(labels)
# Normalized xywh to pixel xyxy format
labels = x.copy()
labels[:, 1] = w * (x[:, 1] - x[:, 3] / 2) + padw
labels[:, 2] = h * (x[:, 2] - x[:, 4] / 2) + padh
labels[:, 3] = w * (x[:, 1] + x[:, 3] / 2) + padw
labels[:, 4] = h * (x[:, 2] + x[:, 4] / 2) + padh
else:
labels = np.zeros((0, 5), dtype=np.float32)
labels4.append(labels)
# Concat/clip labels # Concat/clip labels
if len(labels4): if len(labels4):
@ -585,10 +574,10 @@ def load_mosaic(self, index):
# Augment # Augment
# img4 = img4[s // 2: int(s * 1.5), s // 2:int(s * 1.5)] # center crop (WARNING, requires box pruning) # img4 = img4[s // 2: int(s * 1.5), s // 2:int(s * 1.5)] # center crop (WARNING, requires box pruning)
img4, labels4 = random_affine(img4, labels4, img4, labels4 = random_affine(img4, labels4,
degrees=self.hyp['degrees'] * 1, degrees=self.hyp['degrees'],
translate=self.hyp['translate'] * 1, translate=self.hyp['translate'],
scale=self.hyp['scale'] * 1, scale=self.hyp['scale'],
shear=self.hyp['shear'] * 1, shear=self.hyp['shear'],
border=-s // 2) # border to remove border=-s // 2) # border to remove
return img4, labels4 return img4, labels4
@ -688,7 +677,7 @@ def random_affine(img, targets=(), degrees=10, translate=.1, scale=.1, shear=10,
area = w * h area = w * h
area0 = (targets[:, 3] - targets[:, 1]) * (targets[:, 4] - targets[:, 2]) area0 = (targets[:, 3] - targets[:, 1]) * (targets[:, 4] - targets[:, 2])
ar = np.maximum(w / (h + 1e-16), h / (w + 1e-16)) # aspect ratio ar = np.maximum(w / (h + 1e-16), h / (w + 1e-16)) # aspect ratio
i = (w > 4) & (h > 4) & (area / (area0 + 1e-16) > 0.2) & (ar < 10) i = (w > 4) & (h > 4) & (area / (area0 * s + 1e-16) > 0.2) & (ar < 10)
targets = targets[i] targets = targets[i]
targets[:, 1:5] = xy[i] targets[:, 1:5] = xy[i]