def forward(self, is_train, req, in_data, out_data, aux):
assert self._batch_rois % self._batch_images == 0, \
'BATCHIMAGES {} must devide BATCH_ROIS {}'.format(self._batch_images, self._batch_rois)
rois_per_image = self._batch_rois / self._batch_images
fg_rois_per_image = np.round(self._fg_fraction * rois_per_image).astype(int)
all_rois = in_data[0].asnumpy()
gt_boxes = in_data[1].asnumpy()
# Include ground-truth boxes in the set of candidate rois
zeros = np.zeros((gt_boxes.shape[0], 1), dtype=gt_boxes.dtype)
all_rois = np.vstack((all_rois, np.hstack((zeros, gt_boxes[:, :-1]))))
# Sanity check: single batch only
assert np.all(all_rois[:, 0] == 0), 'Only single item batches are supported'
rois, labels, bbox_targets, bbox_weights = \
sample_rois(all_rois, fg_rois_per_image, rois_per_image, self._num_classes, gt_boxes=gt_boxes)
if logger.level == logging.DEBUG:
logger.debug("labels: %s" % labels)
logger.debug('num fg: {}'.format((labels > 0).sum()))
logger.debug('num bg: {}'.format((labels == 0).sum()))
self._count += 1
self._fg_num += (labels > 0).sum()
self._bg_num += (labels == 0).sum()
logger.debug("self._count: %d" % self._count)
logger.debug('num fg avg: %d' % (self._fg_num / self._count))
logger.debug('num bg avg: %d' % (self._bg_num / self._count))
logger.debug('ratio: %.3f' % (float(self._fg_num) / float(self._bg_num)))
for ind, val in enumerate([rois, labels, bbox_targets, bbox_weights]):
self.assign(out_data[ind], req[ind], val)
def forward(self, is_train, req, in_data, out_data, aux):
# from ipdb import set_trace;set_trace()
assert self._batch_rois % self._batch_images == 0, \
'BATCHIMAGES {} must devide BATCH_ROIS {}'.format(self._batch_images, self._batch_rois)
rois_per_image = self._batch_rois / self._batch_images
fg_rois_per_image = np.round(self._fg_fraction * rois_per_image).astype(np.int)
all_rois = in_data[0].asnumpy()
gt_boxes = in_data[1].asnumpy()
# Include ground-truth boxes in the set of candidate rois
zeros = np.zeros((gt_boxes.shape[0], 1), dtype=gt_boxes.dtype)
all_rois = np.vstack((all_rois, np.hstack((zeros, gt_boxes[:, :-1]))))
# Sanity check: single batch only
assert np.all(all_rois[:, 0] == 0), 'Only single item batches are supported'
rois, labels, bbox_targets, bbox_weights = \
sample_rois(all_rois, fg_rois_per_image, rois_per_image, self._num_classes, gt_boxes=gt_boxes)
if logger.level == logging.DEBUG:
logger.debug("labels: %s" % labels)
logger.debug('num fg: {}'.format((labels > 0).sum()))
logger.debug('num bg: {}'.format((labels == 0).sum()))
self._count += 1
self._fg_num += (labels > 0).sum()
self._bg_num += (labels == 0).sum()
logger.debug("self._count: %d" % self._count)
logger.debug('num fg avg: %d' % (self._fg_num / self._count))
logger.debug('num bg avg: %d' % (self._bg_num / self._count))
logger.debug('ratio: %.3f' % (float(self._fg_num) / float(self._bg_num)))
for ind, val in enumerate([rois, labels, bbox_targets, bbox_weights]):
self.assign(out_data[ind], req[ind], val)
def forward(self, is_train, req, in_data, out_data, aux):
assert self._batch_rois % self._batch_images == 0, \
'BATCHIMAGES {} must devide BATCH_ROIS {}'.format(self._batch_images, self._batch_rois)
rois_per_image = old_div(self._batch_rois, self._batch_images)
fg_rois_per_image = np.round(self._fg_fraction *
rois_per_image).astype(int)
all_rois = in_data[0].asnumpy()
gt_boxes = in_data[1].asnumpy()
# Include ground-truth boxes in the set of candidate rois
zeros = np.zeros((gt_boxes.shape[0], 1), dtype=gt_boxes.dtype)
all_rois = np.vstack((all_rois, np.hstack((zeros, gt_boxes[:, :-1]))))
# Sanity check: single batch only
assert np.all(
all_rois[:, 0] == 0), 'Only single item batches are supported'
rois, labels, bbox_targets, bbox_weights = \
sample_rois(all_rois, fg_rois_per_image, rois_per_image, self._num_classes, gt_boxes=gt_boxes)
if DEBUG:
print("labels=", labels)
print('num fg: {}'.format((labels > 0).sum()))
print('num bg: {}'.format((labels == 0).sum()))
self._count += 1
self._fg_num += (labels > 0).sum()
self._bg_num += (labels == 0).sum()
print("self._count=", self._count)
print('num fg avg: {}'.format(old_div(self._fg_num, self._count)))
print('num bg avg: {}'.format(old_div(self._bg_num, self._count)))
print('ratio: {:.3f}'.format(
old_div(float(self._fg_num), float(self._bg_num))))
for ind, val in enumerate([rois, labels, bbox_targets, bbox_weights]):
self.assign(out_data[ind], req[ind], val)
def forward(self, is_train, req, in_data, out_data, aux):
assert self._batch_rois % self._batch_images == 0, \
'BATCHIMAGES {} must devide BATCH_ROIS {}'.format(self._batch_images, self._batch_rois)
# 每张图片多少个 rois，论文中是64
rois_per_image = self._batch_rois / self._batch_images
# 每张图片的前景(正样本)数量 128*0.25 = 36
fg_rois_per_image = np.round(self._fg_fraction *
rois_per_image).astype(int)
# 所有 rois
all_rois = in_data[0].asnumpy()
# GT boxes (x1, y1, x2, y2, label)
gt_boxes = in_data[1].asnumpy()
# Include ground-truth boxes in the set of candidate rois
zeros = np.zeros((gt_boxes.shape[0], 1), dtype=gt_boxes.dtype)
all_rois = np.vstack((all_rois, np.hstack((zeros, gt_boxes[:, :-1]))))
# Sanity check: single batch only
assert np.all(
all_rois[:, 0] == 0), 'Only single item batches are supported'
# generate random sample of ROIs comprising foreground and background examples
rois, labels, bbox_targets, bbox_weights = \
sample_rois(all_rois, fg_rois_per_image, rois_per_image, self._num_classes, gt_boxes=gt_boxes)
if logger.level == logging.DEBUG:
logger.debug("labels: %s" % labels)
logger.debug('num fg: {}'.format((labels > 0).sum()))
logger.debug('num bg: {}'.format((labels == 0).sum()))
self._count += 1
self._fg_num += (labels > 0).sum()
self._bg_num += (labels == 0).sum()
logger.debug("self._count: %d" % self._count)
logger.debug('num fg avg: %d' % (self._fg_num / self._count))
logger.debug('num bg avg: %d' % (self._bg_num / self._count))
logger.debug('ratio: %.3f' %
(float(self._fg_num) / float(self._bg_num)))
for ind, val in enumerate([rois, labels, bbox_targets, bbox_weights]):
self.assign(out_data[ind], req[ind], val)
