def set_model(self, model_name):
model = FasterRCNN(self.gpu)
model.train = False
serializers.load_npz(model_name, model)
self.model = model
if chainer.cuda.available and self.gpu >= 0:
self.model.to_gpu(self.gpu)
def test_backward_cpu_VGG16(self):
gpu = -1
trunk = VGG16
rpn_in_ch = 512
rpn_out_ch = 512
n_anchors = 9
feat_stride = 16
anchor_scales = [8, 16, 32]
num_classes = 21
spatial_scale = 0.0625
model = FasterRCNN(
gpu, trunk, rpn_in_ch, rpn_out_ch, n_anchors, feat_stride,
anchor_scales, num_classes, spatial_scale)
opt = optimizers.Adam()
opt.setup(model)
model.train = True
rpn_cls_loss, rpn_loss_bbox, loss_bbox, loss_cls = \
model(chainer.Variable(self.x, volatile=False),
self.im_info, self.gt_boxes)
model.zerograds()
rpn_cls_loss.backward()
opt.update()
rpn_cls_loss, rpn_loss_bbox, loss_bbox, loss_cls = \
model(chainer.Variable(self.x, volatile=False),
self.im_info, self.gt_boxes)
model.zerograds()
rpn_loss_bbox.backward()
opt.update()
rpn_cls_loss, rpn_loss_bbox, loss_bbox, loss_cls = \
model(chainer.Variable(self.x, volatile=False),
self.im_info, self.gt_boxes)
model.zerograds()
loss_bbox.backward()
opt.update()
rpn_cls_loss, rpn_loss_bbox, loss_bbox, loss_cls = \
model(chainer.Variable(self.x, volatile=False),
self.im_info, self.gt_boxes)
model.zerograds()
loss_cls.backward()
opt.update()
def test_forward_cpu_VGG16(self):
print('test_forward_cpu_VGG16')
gpu = -1
trunk = VGG16
rpn_in_ch = 512
rpn_out_ch = 512
n_anchors = 9
feat_stride = 16
anchor_scales = [8, 16, 32]
num_classes = 21
spatial_scale = 0.0625
model = FasterRCNN(gpu, trunk, rpn_in_ch, rpn_out_ch, n_anchors,
feat_stride, anchor_scales, num_classes,
spatial_scale)
model.train = False
ret = model(chainer.Variable(self.x, volatile=True), self.im_info)
assert (len(ret) == 2)
assert (isinstance(ret[0], chainer.Variable))
assert (isinstance(ret[1], np.ndarray))
def test_forward_cpu_VGG16(self):
print('test_forward_cpu_VGG16')
gpu = -1
trunk = VGG16
rpn_in_ch = 512
rpn_out_ch = 512
n_anchors = 9
feat_stride = 16
anchor_scales = [8, 16, 32]
num_classes = 21
spatial_scale = 0.0625
model = FasterRCNN(
gpu, trunk, rpn_in_ch, rpn_out_ch, n_anchors, feat_stride,
anchor_scales, num_classes, spatial_scale)
model.train = False
ret = model(chainer.Variable(self.x, volatile=True), self.im_info)
assert(len(ret) == 2)
assert(isinstance(ret[0], chainer.Variable))
assert(isinstance(ret[1], np.ndarray))
def get_model(gpu):
model = FasterRCNN(gpu)
model.train = False
serializers.load_npz('VGG16_faster_rcnn_final.model', model)
return model
def get_model(gpu):
model = FasterRCNN(gpu)
model.train = False
serializers.load_npz('./VGG16.model', model)
return model
def get_model(gpu):
model = FasterRCNN(gpu)
model.train = False
serializers.load_npz('data/VGG16_faster_rcnn_final.model', model)
return model
import sys
sys.path.insert(0, './lib/models')
sys.path.insert(0, '/media/haria/data/py-faster-rcnn/lib')
sys.path.insert(0, '/media/haria/data/py-faster-rcnn/caffe-fast-rcnn/python')
import caffe
from lib.models.faster_rcnn import FasterRCNN
import cPickle as pickle
from lib.models.npz import load_npz
from chainer import serializers
param_fn = './VGG16_faster_rcnn_final.caffemodel'
model_fn = './test.prototxt'
vgg = FasterRCNN()
net = caffe.Net(model_fn, param_fn, caffe.TEST)
for name, param in net.params.iteritems():
name = name.replace("/", "_")
if name not in [
"rpn_cls_score", "rpn_bbox_pred", "fc6", "fc7", "cls_score",
"bbox_pred"
]:
trunk = vgg.trunk
layer = getattr(trunk, name)
layer.W = param[0].data
layer.b = param[1].data
setattr(trunk, name, layer)
else:
layer = getattr(vgg, name)
def test_backward_cpu_VGG16(self):
gpu = -1
trunk = VGG16
rpn_in_ch = 512
rpn_out_ch = 512
n_anchors = 9
feat_stride = 16
anchor_scales = [8, 16, 32]
num_classes = 21
spatial_scale = 0.0625
model = FasterRCNN(gpu, trunk, rpn_in_ch, rpn_out_ch, n_anchors,
feat_stride, anchor_scales, num_classes,
spatial_scale)
opt = optimizers.Adam()
opt.setup(model)
model.train = True
rpn_cls_loss, rpn_loss_bbox, loss_bbox, loss_cls = \
model(chainer.Variable(self.x, volatile=False),
self.im_info, self.gt_boxes)
model.zerograds()
rpn_cls_loss.backward()
opt.update()
rpn_cls_loss, rpn_loss_bbox, loss_bbox, loss_cls = \
model(chainer.Variable(self.x, volatile=False),
self.im_info, self.gt_boxes)
model.zerograds()
rpn_loss_bbox.backward()
opt.update()
rpn_cls_loss, rpn_loss_bbox, loss_bbox, loss_cls = \
model(chainer.Variable(self.x, volatile=False),
self.im_info, self.gt_boxes)
model.zerograds()
loss_bbox.backward()
opt.update()
rpn_cls_loss, rpn_loss_bbox, loss_bbox, loss_cls = \
model(chainer.Variable(self.x, volatile=False),
self.im_info, self.gt_boxes)
model.zerograds()
loss_cls.backward()
opt.update()
