def forward(self, bottom, top):
"""Compute loss, select RoIs using OHEM. Use RoIs to get blobs and copy them into this layer's top blob vector."""
cls_prob = bottom[0].data
bbox_pred = bottom[1].data
rois = bottom[2].data
labels = bottom[3].data
labels = labels.astype(int)
if cfg.TRAIN.BBOX_REG:
bbox_target = bottom[4].data
bbox_inside_weights = bottom[5].data
bbox_outside_weights = bottom[6].data
im_data = bottom[7].data # Alice: 2017.11.16 adding roi visualization
else:
im_data = bottom[4].data # Alice: 2017.11.16 adding roi visualization
bbox_target = None
bbox_inside_weights = None
bbox_outside_weights = None
flt_min = np.finfo(float).eps
# classification loss
loss = [ -1 * np.log(max(x, flt_min)) \
for x in [cls_prob[i,label] for i, label in enumerate(labels)]]
if cfg.TRAIN.BBOX_REG:
# bounding-box regression loss
# d := w * (b0 - b1)
# smoothL1(x) = 0.5 * x^2 if |x| < 1
# |x| - 0.5 otherwise
def smoothL1(x):
if abs(x) < 1:
return 0.5 * x * x
else:
return abs(x) - 0.5
bbox_loss = np.zeros(labels.shape[0])
for i in np.where(labels > 0 )[0]:
indices = np.where(bbox_inside_weights[i,:] != 0)[0]
bbox_loss[i] = sum(bbox_outside_weights[i,indices] * [smoothL1(x) \
for x in bbox_inside_weights[i,indices] * (bbox_pred[i,indices] - bbox_target[i,indices])])
loss += bbox_loss
#original version(don't need to visualization)
#blobs = get_ohem_minibatch(loss, rois, labels, bbox_target, \
#bbox_inside_weights, bbox_outside_weights)
#Alice: 2017.11.16 (adding bbox visualization, one more parameter)
blobs = get_ohem_minibatch(im_data, loss, rois, labels, bbox_target, \
bbox_inside_weights, bbox_outside_weights)
for blob_name, blob in blobs.iteritems():
top_ind = self._name_to_top_map[blob_name]
# Reshape net's input blobs
top[top_ind].reshape(*(blob.shape))
# Copy data into net's input blobs
top[top_ind].data[...] = blob.astype(np.float32, copy=False)
def forward(self, bottom, top):
"""Compute loss, select RoIs using OHEM. Use RoIs to get blobs and copy them into this layer's top blob vector."""
cls_prob = bottom[0].data
bbox_pred = bottom[1].data
rois = bottom[2].data
labels = bottom[3].data
if cfg.TRAIN.BBOX_REG:
bbox_target = bottom[4].data
bbox_inside_weights = bottom[5].data
bbox_outside_weights = bottom[6].data
else:
bbox_target = None
bbox_inside_weights = None
bbox_outside_weights = None
flt_min = np.finfo(float).eps
# classification loss
loss = [ -1 * np.log(max(x, flt_min)) \
for x in [cls_prob[i,label] for i, label in enumerate(labels)]]
if cfg.TRAIN.BBOX_REG:
# bounding-box regression loss
# d := w * (b0 - b1)
# smoothL1(x) = 0.5 * x^2 if |x| < 1
# |x| - 0.5 otherwise
def smoothL1(x):
if abs(x) < 1:
return 0.5 * x * x
else:
return abs(x) - 0.5
bbox_loss = np.zeros(labels.shape[0])
for i in np.where(labels > 0 )[0]:
indices = np.where(bbox_inside_weights[i,:] != 0)[0]
bbox_loss[i] = sum(bbox_outside_weights[i,indices] * [smoothL1(x) \
for x in bbox_inside_weights[i,indices] * (bbox_pred[i,indices] - bbox_target[i,indices])])
loss += bbox_loss
blobs = get_ohem_minibatch(loss, rois, labels, bbox_target, \
bbox_inside_weights, bbox_outside_weights)
for blob_name, blob in blobs.iteritems():
top_ind = self._name_to_top_map[blob_name]
# Reshape net's input blobs
top[top_ind].reshape(*(blob.shape))
# Copy data into net's input blobs
top[top_ind].data[...] = blob.astype(np.float32, copy=False)
def forward(self, bottom, top):
"""Compute loss, select RoIs using OHEM. Use RoIs to get blobs and copy them into this layer's top blob vector."""
cls_prob = bottom[0].data
bbox_pred = bottom[1].data
rois = bottom[2].data
labels = bottom[3].data
self._count_iter = (self._count_iter + 1) % self._iter_size
if cfg.TRAIN.BBOX_REG:
bbox_target = bottom[4].data
bbox_inside_weights = bottom[5].data
bbox_outside_weights = bottom[6].data
else:
bbox_target = None
bbox_inside_weights = None
bbox_outside_weights = None
flt_min = np.finfo(float).eps
# classification loss
loss = [ -1 * np.log(max(x, flt_min)) \
for x in [cls_prob[i,label] for i, label in enumerate(labels)]]
if cfg.TRAIN.BBOX_REG:
# bounding-box regression loss
# d := w * (b0 - b1)
# smoothL1(x) = 0.5 * x^2 if |x| < 1
# |x| - 0.5 otherwise
def smoothL1(x):
if abs(x) < 1:
return 0.5 * x * x
else:
return abs(x) - 0.5
bbox_loss = np.zeros(labels.shape[0])
for i in np.where(labels > 0 )[0]:
indices = np.where(bbox_inside_weights[i,:] != 0)[0]
bbox_loss[i] = sum(bbox_outside_weights[i,indices] * [smoothL1(x) \
for x in bbox_inside_weights[i,indices] * (bbox_pred[i,indices] - bbox_target[i,indices])])
loss += bbox_loss
blobs = []
hard_inds = []
if self._count_iter < self._maintain_before or cfg.TRAIN.OHEM_RATIO < 0.1:
blobs, hard_inds = get_ohem_minibatch(loss, rois, labels, bbox_target, bbox_inside_weights, bbox_outside_weights)
else:
blobs, hard_inds = get_ohem_minibatch_ratio(loss, rois, labels, bbox_target, bbox_inside_weights, bbox_outside_weights, ratio=cfg.TRAIN.OHEM_RATIO, hard_negative=cfg.TRAIN.OHEM_HARD_NEG)
for blob_name, blob in blobs.iteritems():
top_ind = self._name_to_top_map[blob_name]
# Reshape net's input blobs
top[top_ind].reshape(*(blob.shape))
# Copy data into net's input blobs
top[top_ind].data[...] = blob.astype(np.float32, copy=False)
# used for ASDN
if cfg.TRAIN.USE_ASDN:
prop_before = cls_prob[hard_inds]
top_ind = self._name_to_top_map['prop_before']
top[top_ind].reshape(*(prop_before.shape))
top[top_ind].data[...] = prop_before.astype(np.float32, copy=False)