def build_program(main_prog, startup_prog, args, data_args):
image_shape = [3, data_args.resize_h, data_args.resize_w]
if 'coco' in data_args.dataset:
num_classes = 91
elif 'pascalvoc' in data_args.dataset:
num_classes = 21
with fluid.program_guard(main_prog, startup_prog):
py_reader = fluid.layers.py_reader(
capacity=64,
shapes=[[-1] + image_shape, [-1, 4], [-1, 1], [-1, 1]],
lod_levels=[0, 1, 1, 1],
dtypes=["float32", "float32", "int32", "int32"],
use_double_buffer=True)
with fluid.unique_name.guard():
image, gt_box, gt_label, difficult = fluid.layers.read_file(
py_reader)
locs, confs, box, box_var = mobile_net(num_classes, image,
image_shape)
nmsed_out = fluid.layers.detection_output(
locs, confs, box, box_var, nms_threshold=args.nms_threshold)
with fluid.program_guard(main_prog):
map = fluid.metrics.DetectionMAP(nmsed_out,
gt_label,
gt_box,
difficult,
num_classes,
overlap_threshold=0.5,
evaluate_difficult=False,
ap_version=args.ap_version)
return py_reader, map
def infer(args, data_args, image_path, model_dir):
image_shape = [3, data_args.resize_h, data_args.resize_w]
num_classes = 2
label_list = data_args.label_list
image = fluid.layers.data(name='image', shape=image_shape, dtype='float32')
locs, confs, box, box_var = mobile_net(num_classes, image, image_shape)
nmsed_out = fluid.layers.detection_output(locs,
confs,
box,
box_var,
nms_threshold=args.nms_threshold)
place = fluid.CUDAPlace(0) if args.use_gpu else fluid.CPUPlace()
exe = fluid.Executor(place)
# yapf: disable
if model_dir:
def if_exist(var):
return os.path.exists(os.path.join(model_dir, var.name))
fluid.io.load_vars(exe, model_dir, predicate=if_exist)
# yapf: enable
test_images = os.listdir(image_path)
test_images.remove('label_list')
test_images = sorted(test_images, key=lambda x: int(x[:-4]))
for test_image in test_images:
test_path = os.path.join(image_path, test_image)
infer_reader = reader.infer(data_args, test_path)
feeder = fluid.DataFeeder(place=place, feed_list=[image])
data = infer_reader()
# switch network to test mode (i.e. batch norm test mode)
test_program = fluid.default_main_program().clone(for_test=True)
nmsed_out_v, = exe.run(test_program,
feed=feeder.feed([[data]]),
fetch_list=[nmsed_out],
return_numpy=False)
nmsed_out_v = np.array(nmsed_out_v)
count = 0
for dt in nmsed_out_v:
category_id, score, xmin, ymin, xmax, ymax = dt.tolist()
if score < args.confs_threshold:
continue
count += 1
#draw_bounding_box_on_image(test_path, nmsed_out_v, args.confs_threshold,
# label_list)
with open('result.csv', 'a') as f:
test_path = test_path.split('/')[-1]
f.write(test_path[:-4] + ',' + str(count) + '\n')
def infer(args, data_args, image_path, model_dir):
image_shape = [3, data_args.resize_h, data_args.resize_w]
if 'coco' in data_args.dataset:
num_classes = 91
# cocoapi
from pycocotools.coco import COCO
from pycocotools.cocoeval import COCOeval
label_fpath = os.path.join(data_dir, label_file)
coco = COCO(label_fpath)
category_ids = coco.getCatIds()
label_list = {
item['id']: item['name']
for item in coco.loadCats(category_ids)
}
label_list[0] = ['background']
elif 'pascalvoc' in data_args.dataset:
num_classes = 21
label_list = data_args.label_list
image = fluid.layers.data(name='image', shape=image_shape, dtype='float32')
locs, confs, box, box_var = mobile_net(num_classes, image, image_shape)
nmsed_out = fluid.layers.detection_output(locs,
confs,
box,
box_var,
nms_threshold=args.nms_threshold)
place = fluid.CUDAPlace(0) if args.use_gpu else fluid.CPUPlace()
exe = fluid.Executor(place)
# yapf: disable
if model_dir:
def if_exist(var):
return os.path.exists(os.path.join(model_dir, var.name))
fluid.io.load_vars(exe, model_dir, predicate=if_exist)
# yapf: enable
infer_reader = reader.infer(data_args, image_path)
feeder = fluid.DataFeeder(place=place, feed_list=[image])
data = infer_reader()
# switch network to test mode (i.e. batch norm test mode)
test_program = fluid.default_main_program().clone(for_test=True)
nmsed_out_v, = exe.run(test_program,
feed=feeder.feed([[data]]),
fetch_list=[nmsed_out],
return_numpy=False)
nmsed_out_v = np.array(nmsed_out_v)
draw_bounding_box_on_image(image_path, nmsed_out_v, args.confs_threshold,
label_list)
def build_program(main_prog, startup_prog, train_params, is_train):
image_shape = train_params['image_shape']
class_num = train_params['class_num']
ap_version = train_params['ap_version']
outs = []
with fluid.program_guard(main_prog, startup_prog):
py_reader = fluid.layers.py_reader(
capacity=64,
shapes=[[-1] + image_shape, [-1, 4], [-1, 1], [-1, 1]],
lod_levels=[0, 1, 1, 1],
dtypes=["float32", "float32", "int32", "int32"],
use_double_buffer=True)
with fluid.unique_name.guard():
image, gt_box, gt_label, difficult = fluid.layers.read_file(
py_reader)
locs, confs, box, box_var = mobile_net(class_num, image,
image_shape)
if is_train:
with fluid.unique_name.guard("train"):
loss = fluid.layers.ssd_loss(locs, confs, gt_box, gt_label,
box, box_var)
loss = fluid.layers.reduce_sum(loss)
optimizer = optimizer_setting(train_params)
optimizer.minimize(loss)
outs = [py_reader, loss]
else:
with fluid.unique_name.guard("inference"):
nmsed_out = fluid.layers.detection_output(
locs, confs, box, box_var, nms_threshold=0.45)
map_eval = fluid.metrics.DetectionMAP(
nmsed_out,
gt_label,
gt_box,
difficult,
class_num,
overlap_threshold=0.5,
evaluate_difficult=False,
ap_version=ap_version)
# nmsed_out and image is used to save mode for inference
outs = [py_reader, map_eval, nmsed_out, image]
return outs
def train(args,
train_file_list,
val_file_list,
data_args,
learning_rate,
batch_size,
num_passes,
model_save_dir,
pretrained_model=None):
image_shape = [3, data_args.resize_h, data_args.resize_w]
num_classes = 2
devices = os.getenv("CUDA_VISIBLE_DEVICES") or ""
devices_num = len(devices.split(","))
image = fluid.layers.data(name='image', shape=image_shape, dtype='float32')
gt_box = fluid.layers.data(name='gt_box',
shape=[4],
dtype='float32',
lod_level=1)
gt_label = fluid.layers.data(name='gt_label',
shape=[1],
dtype='int32',
lod_level=1)
difficult = fluid.layers.data(name='gt_difficult',
shape=[1],
dtype='int32',
lod_level=1)
locs, confs, box, box_var = mobile_net(num_classes, image, image_shape)
nmsed_out = fluid.layers.detection_output(locs,
confs,
box,
box_var,
nms_threshold=args.nms_threshold)
loss = fluid.layers.ssd_loss(locs, confs, gt_box, gt_label, box, box_var)
loss = fluid.layers.reduce_sum(loss)
test_program = fluid.default_main_program().clone(for_test=True)
with fluid.program_guard(test_program):
map_eval = fluid.evaluator.DetectionMAP(nmsed_out,
gt_label,
gt_box,
difficult,
num_classes,
overlap_threshold=0.5,
evaluate_difficult=False,
ap_version=args.ap_version)
epocs = 4800 / batch_size
boundaries = [epocs * 40, epocs * 60, epocs * 80, epocs * 100]
values = [
learning_rate, learning_rate * 0.5, learning_rate * 0.25,
learning_rate * 0.1, learning_rate * 0.01
]
optimizer = fluid.optimizer.RMSProp(
learning_rate=fluid.layers.piecewise_decay(boundaries, values),
regularization=fluid.regularizer.L2Decay(0.00005),
)
optimizer.minimize(loss)
place = fluid.CUDAPlace(0) if args.use_gpu else fluid.CPUPlace()
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())
if pretrained_model:
def if_exist(var):
return os.path.exists(os.path.join(pretrained_model, var.name))
fluid.io.load_vars(exe, pretrained_model, predicate=if_exist)
if args.parallel:
train_exe = fluid.ParallelExecutor(use_cuda=args.use_gpu,
loss_name=loss.name)
train_reader = paddle.batch(reader.train(data_args, train_file_list),
batch_size=batch_size)
test_reader = paddle.batch(reader.test(data_args, val_file_list),
batch_size=batch_size)
feeder = fluid.DataFeeder(place=place,
feed_list=[image, gt_box, gt_label, difficult])
def save_model(postfix):
model_path = os.path.join(model_save_dir, postfix)
if os.path.isdir(model_path):
shutil.rmtree(model_path)
print('save models to %s' % (model_path))
fluid.io.save_persistables(exe, model_path)
best_map = 0.
def test(pass_id, best_map):
_, accum_map = map_eval.get_map_var()
map_eval.reset(exe)
for batch_id, data in enumerate(test_reader()):
test_map, = exe.run(test_program,
feed=feeder.feed(data),
fetch_list=[accum_map])
if batch_id % 20 == 0:
print("Batch {0}, map {1}".format(batch_id, test_map))
if test_map[0] > best_map:
best_map = test_map[0]
save_model('best_model')
print("Pass {0}, test map {1}".format(pass_id, test_map))
return best_map
train_num = 0
total_train_time = 0.0
for pass_id in range(num_passes):
start_time = time.time()
prev_start_time = start_time
# end_time = 0
every_pass_loss = []
iter = 0
pass_duration = 0.0
for batch_id, data in enumerate(train_reader()):
prev_start_time = start_time
start_time = time.time()
if args.for_model_ce and iter == args.iterations:
break
if len(data) < (devices_num * 2):
print("There are too few data to train on all devices.")
continue
if args.parallel:
loss_v, = train_exe.run(fetch_list=[loss.name],
feed=feeder.feed(data))
else:
loss_v, = exe.run(fluid.default_main_program(),
feed=feeder.feed(data),
fetch_list=[loss])
# end_time = time.time()
loss_v = np.mean(np.array(loss_v))
if batch_id % 20 == 0:
print("Pass {0}, batch {1}, loss {2}, time {3}".format(
pass_id, batch_id, loss_v, start_time - prev_start_time))
if args.for_model_ce and iter >= args.skip_batch_num or pass_id != 0:
batch_duration = time.time() - start_time
pass_duration += batch_duration
train_num += len(data)
every_pass_loss.append(loss_v)
iter += 1
total_train_time += pass_duration
if args.for_model_ce and pass_id == num_passes - 1:
examples_per_sec = train_num / total_train_time
cost = np.mean(every_pass_loss)
with open("train_speed_factor.txt", 'w') as f:
f.write('{:f}\n'.format(examples_per_sec))
with open("train_cost_factor.txt", 'a+') as f:
f.write('{:f}\n'.format(cost))
best_map = test(pass_id, best_map)
if pass_id % 10 == 0 or pass_id == num_passes - 1:
save_model(str(pass_id))
print("Best test map {0}".format(best_map))
def parallel_do(args,
train_file_list,
val_file_list,
data_args,
learning_rate,
batch_size,
num_passes,
model_save_dir,
pretrained_model=None):
image_shape = [3, data_args.resize_h, data_args.resize_w]
if data_args.dataset == 'coco':
num_classes = 81
elif data_args.dataset == 'pascalvoc':
num_classes = 21
image = fluid.layers.data(name='image', shape=image_shape, dtype='float32')
gt_box = fluid.layers.data(name='gt_box',
shape=[4],
dtype='float32',
lod_level=1)
gt_label = fluid.layers.data(name='gt_label',
shape=[1],
dtype='int32',
lod_level=1)
difficult = fluid.layers.data(name='gt_difficult',
shape=[1],
dtype='int32',
lod_level=1)
if args.parallel:
places = fluid.layers.get_places()
pd = fluid.layers.ParallelDo(places, use_nccl=args.use_nccl)
with pd.do():
image_ = pd.read_input(image)
gt_box_ = pd.read_input(gt_box)
gt_label_ = pd.read_input(gt_label)
difficult_ = pd.read_input(difficult)
locs, confs, box, box_var = mobile_net(num_classes, image_,
image_shape)
loss = fluid.layers.ssd_loss(locs, confs, gt_box_, gt_label_, box,
box_var)
nmsed_out = fluid.layers.detection_output(locs,
confs,
box,
box_var,
nms_threshold=0.45)
loss = fluid.layers.reduce_sum(loss)
pd.write_output(loss)
pd.write_output(nmsed_out)
loss, nmsed_out = pd()
loss = fluid.layers.mean(loss)
else:
locs, confs, box, box_var = mobile_net(num_classes, image, image_shape)
nmsed_out = fluid.layers.detection_output(locs,
confs,
box,
box_var,
nms_threshold=0.45)
loss = fluid.layers.ssd_loss(locs, confs, gt_box, gt_label, box,
box_var)
loss = fluid.layers.reduce_sum(loss)
test_program = fluid.default_main_program().clone(for_test=True)
with fluid.program_guard(test_program):
map_eval = fluid.evaluator.DetectionMAP(nmsed_out,
gt_label,
gt_box,
difficult,
num_classes,
overlap_threshold=0.5,
evaluate_difficult=False,
ap_version=args.ap_version)
if data_args.dataset == 'coco':
# learning rate decay in 12, 19 pass, respectively
if '2014' in train_file_list:
boundaries = [82783 / batch_size * 12, 82783 / batch_size * 19]
elif '2017' in train_file_list:
boundaries = [118287 / batch_size * 12, 118287 / batch_size * 19]
elif data_args.dataset == 'pascalvoc':
boundaries = [40000, 60000]
values = [learning_rate, learning_rate * 0.5, learning_rate * 0.25]
optimizer = fluid.optimizer.RMSProp(
learning_rate=fluid.layers.piecewise_decay(boundaries, values),
regularization=fluid.regularizer.L2Decay(0.00005),
)
optimizer.minimize(loss)
place = fluid.CUDAPlace(0) if args.use_gpu else fluid.CPUPlace()
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())
if pretrained_model:
def if_exist(var):
return os.path.exists(os.path.join(pretrained_model, var.name))
fluid.io.load_vars(exe, pretrained_model, predicate=if_exist)
train_reader = paddle.batch(reader.train(data_args, train_file_list),
batch_size=batch_size)
test_reader = paddle.batch(reader.test(data_args, val_file_list),
batch_size=batch_size)
feeder = fluid.DataFeeder(place=place,
feed_list=[image, gt_box, gt_label, difficult])
def test(pass_id):
_, accum_map = map_eval.get_map_var()
map_eval.reset(exe)
test_map = None
for data in test_reader():
test_map = exe.run(test_program,
feed=feeder.feed(data),
fetch_list=[accum_map])
print("Test {0}, map {1}".format(pass_id, test_map[0]))
for pass_id in range(num_passes):
start_time = time.time()
prev_start_time = start_time
end_time = 0
for batch_id, data in enumerate(train_reader()):
prev_start_time = start_time
start_time = time.time()
loss_v = exe.run(fluid.default_main_program(),
feed=feeder.feed(data),
fetch_list=[loss])
end_time = time.time()
if batch_id % 20 == 0:
print("Pass {0}, batch {1}, loss {2}, time {3}".format(
pass_id, batch_id, loss_v[0],
start_time - prev_start_time))
test(pass_id)
if pass_id % 10 == 0 or pass_id == num_passes - 1:
model_path = os.path.join(model_save_dir, str(pass_id))
print 'save models to %s' % (model_path)
fluid.io.save_persistables(exe, model_path)
def parallel_exe(args,
train_file_list,
val_file_list,
data_args,
learning_rate,
batch_size,
num_passes,
model_save_dir='model',
pretrained_model=None):
image_shape = [3, data_args.resize_h, data_args.resize_w]
if data_args.dataset == 'coco':
num_classes = 81
elif data_args.dataset == 'pascalvoc':
num_classes = 21
devices = os.getenv("CUDA_VISIBLE_DEVICES") or ""
devices_num = len(devices.split(","))
image = fluid.layers.data(name='image', shape=image_shape, dtype='float32')
gt_box = fluid.layers.data(name='gt_box',
shape=[4],
dtype='float32',
lod_level=1)
gt_label = fluid.layers.data(name='gt_label',
shape=[1],
dtype='int32',
lod_level=1)
difficult = fluid.layers.data(name='gt_difficult',
shape=[1],
dtype='int32',
lod_level=1)
locs, confs, box, box_var = mobile_net(num_classes, image, image_shape)
nmsed_out = fluid.layers.detection_output(locs,
confs,
box,
box_var,
nms_threshold=0.45)
loss = fluid.layers.ssd_loss(locs, confs, gt_box, gt_label, box, box_var)
loss = fluid.layers.reduce_sum(loss)
test_program = fluid.default_main_program().clone(for_test=True)
with fluid.program_guard(test_program):
map_eval = fluid.evaluator.DetectionMAP(nmsed_out,
gt_label,
gt_box,
difficult,
num_classes,
overlap_threshold=0.5,
evaluate_difficult=False,
ap_version=args.ap_version)
if data_args.dataset == 'coco':
# learning rate decay in 12, 19 pass, respectively
if '2014' in train_file_list:
epocs = 82783 / batch_size
boundaries = [epocs * 12, epocs * 19]
elif '2017' in train_file_list:
epocs = 118287 / batch_size
boundaries = [epcos * 12, epocs * 19]
elif data_args.dataset == 'pascalvoc':
epocs = 19200 / batch_size
boundaries = [epocs * 40, epocs * 60, epocs * 80, epocs * 100]
values = [
learning_rate, learning_rate * 0.5, learning_rate * 0.25,
learning_rate * 0.1, learning_rate * 0.01
]
optimizer = fluid.optimizer.RMSProp(
learning_rate=fluid.layers.piecewise_decay(boundaries, values),
regularization=fluid.regularizer.L2Decay(0.00005),
)
optimizer.minimize(loss)
place = fluid.CUDAPlace(0) if args.use_gpu else fluid.CPUPlace()
exe = fluid.Executor(place)
fluid.default_startup_program.random_seed = 1000
exe.run(fluid.default_startup_program())
if pretrained_model:
def if_exist(var):
return os.path.exists(os.path.join(pretrained_model, var.name))
fluid.io.load_vars(exe, pretrained_model, predicate=if_exist)
if args.parallel:
train_exe = fluid.ParallelExecutor(use_cuda=args.use_gpu,
loss_name=loss.name)
train_reader = paddle.batch(reader.train(data_args, train_file_list),
batch_size=batch_size)
test_reader = paddle.batch(reader.test(data_args, val_file_list),
batch_size=batch_size)
feeder = fluid.DataFeeder(place=place,
feed_list=[image, gt_box, gt_label, difficult])
def save_model(postfix):
model_path = os.path.join(model_save_dir, postfix)
if os.path.isdir(model_path):
shutil.rmtree(model_path)
print 'save models to %s' % (model_path)
fluid.io.save_persistables(exe, model_path)
best_map = 0.
def test(pass_id, best_map):
_, accum_map = map_eval.get_map_var()
map_eval.reset(exe)
test_map = None
for data in test_reader():
test_map = exe.run(test_program,
feed=feeder.feed(data),
fetch_list=[accum_map])
if test_map[0] > best_map:
best_map = test_map[0]
save_model('best_model')
print("Test {0}, map {1}".format(pass_id, test_map[0]))
train_num = 0
total_train_time = 0.0
total_iters = 0
for pass_id in range(num_passes):
every_pass_loss = []
iter = 0
pass_duration = 0.0
for batch_id, data in enumerate(train_reader()):
batch_start = time.time()
if iter == args.iterations:
break
if len(data) < devices_num: continue
if args.parallel:
loss_v, = train_exe.run(fetch_list=[loss.name],
feed=feeder.feed(data))
else:
loss_v, = exe.run(fluid.default_main_program(),
feed=feeder.feed(data),
fetch_list=[loss])
loss_v = np.mean(np.array(loss_v))
if batch_id % 20 == 0:
print("Pass {0}, batch {1}, loss {2}, time {3}".format(
pass_id, batch_id, loss_v,
time.time() - batch_start))
if iter >= args.skip_batch_num or pass_id != 0:
batch_duration = time.time() - batch_start
pass_duration += batch_duration
train_num += len(data)
every_pass_loss.append(loss_v)
iter += 1
total_iters += 1
#test(pass_id, best_map)
total_train_time += pass_duration
print("Pass:%d, Loss:%f, Handle Images Duration: %f\n" %
(pass_id, np.mean(every_pass_loss), pass_duration))
if pass_id == num_passes - 1:
examples_per_sec = train_num / total_train_time
train_cost_kpi.add_record(np.mean(every_pass_loss))
train_speed_kpi.add_record(
np.array(examples_per_sec, dtype='float'))
four_card_speed_kpi.add_record(
np.array(examples_per_sec, dtype='float'))
if args.gpu_card_num == 1:
train_cost_kpi.persist()
train_speed_kpi.persist()
else:
four_card_speed_kpi.persist()
print("Best test map {0}".format(best_map))
def eval(args, data_args, test_list, batch_size, model_dir=None):
image_shape = [3, data_args.resize_h, data_args.resize_w]
num_classes = 2
image = fluid.layers.data(name='image', shape=image_shape, dtype='float32')
gt_box = fluid.layers.data(name='gt_box',
shape=[4],
dtype='float32',
lod_level=1)
gt_label = fluid.layers.data(name='gt_label',
shape=[1],
dtype='int32',
lod_level=1)
difficult = fluid.layers.data(name='gt_difficult',
shape=[1],
dtype='int32',
lod_level=1)
locs, confs, box, box_var = mobile_net(num_classes, image, image_shape)
nmsed_out = fluid.layers.detection_output(locs,
confs,
box,
box_var,
nms_threshold=args.nms_threshold)
loss = fluid.layers.ssd_loss(locs, confs, gt_box, gt_label, box, box_var)
loss = fluid.layers.reduce_sum(loss)
place = fluid.CUDAPlace(0) if args.use_gpu else fluid.CPUPlace()
exe = fluid.Executor(place)
# yapf: disable
if model_dir:
def if_exist(var):
return os.path.exists(os.path.join(model_dir, var.name))
fluid.io.load_vars(exe, model_dir, predicate=if_exist)
# yapf: enable
test_reader = paddle.batch(reader.test(data_args, test_list),
batch_size=batch_size)
feeder = fluid.DataFeeder(place=place,
feed_list=[image, gt_box, gt_label, difficult])
def test():
# switch network to test mode (i.e. batch norm test mode)
test_program = fluid.default_main_program().clone(for_test=True)
with fluid.program_guard(test_program):
map_eval = fluid.evaluator.DetectionMAP(nmsed_out,
gt_label,
gt_box,
difficult,
num_classes,
overlap_threshold=0.5,
evaluate_difficult=False,
ap_version=args.ap_version)
_, accum_map = map_eval.get_map_var()
map_eval.reset(exe)
for batch_id, data in enumerate(test_reader()):
test_map, = exe.run(test_program,
feed=feeder.feed(data),
fetch_list=[accum_map])
if batch_id % 20 == 0:
print("Batch {0}, map {1}".format(batch_id, test_map))
print("Test model {0}, map {1}".format(model_dir, test_map))
test()
def eval(args, data_args, test_list, batch_size, model_dir=None):
image_shape = [3, data_args.resize_h, data_args.resize_w]
num_classes = 91
image = fluid.layers.data(name='image', shape=image_shape, dtype='float32')
gt_box = fluid.layers.data(name='gt_box',
shape=[4],
dtype='float32',
lod_level=1)
gt_label = fluid.layers.data(name='gt_label',
shape=[1],
dtype='int32',
lod_level=1)
gt_iscrowd = fluid.layers.data(name='gt_iscrowd',
shape=[1],
dtype='int32',
lod_level=1)
gt_image_info = fluid.layers.data(name='gt_image_id',
shape=[3],
dtype='int32')
locs, confs, box, box_var = mobile_net(num_classes, image, image_shape)
nmsed_out = fluid.layers.detection_output(locs,
confs,
box,
box_var,
nms_threshold=args.nms_threshold)
loss = fluid.layers.ssd_loss(locs, confs, gt_box, gt_label, box, box_var)
loss = fluid.layers.reduce_sum(loss)
place = fluid.CUDAPlace(0) if args.use_gpu else fluid.CPUPlace()
exe = fluid.Executor(place)
exe.run(fluid.default_startup_program())
# yapf: disable
if model_dir:
def if_exist(var):
return os.path.exists(os.path.join(model_dir, var.name))
fluid.io.load_vars(exe, model_dir, predicate=if_exist)
# yapf: enable
test_reader = reader.test(data_args, test_list, batch_size)
feeder = fluid.DataFeeder(
place=place,
feed_list=[image, gt_box, gt_label, gt_iscrowd, gt_image_info])
def get_dt_res(nmsed_out_v, data):
dts_res = []
lod = nmsed_out_v[0].lod()[0]
nmsed_out_v = np.array(nmsed_out_v[0])
real_batch_size = min(batch_size, len(data))
assert (len(lod) == real_batch_size + 1), \
"Error Lod Tensor offset dimension. Lod({}) vs. batch_size({})".format(len(lod), batch_size)
k = 0
for i in range(real_batch_size):
dt_num_this_img = lod[i + 1] - lod[i]
image_id = int(data[i][4][0])
image_width = int(data[i][4][1])
image_height = int(data[i][4][2])
for j in range(dt_num_this_img):
dt = nmsed_out_v[k]
k = k + 1
category_id, score, xmin, ymin, xmax, ymax = dt.tolist()
xmin = max(min(xmin, 1.0), 0.0) * image_width
ymin = max(min(ymin, 1.0), 0.0) * image_height
xmax = max(min(xmax, 1.0), 0.0) * image_width
ymax = max(min(ymax, 1.0), 0.0) * image_height
w = xmax - xmin
h = ymax - ymin
bbox = [xmin, ymin, w, h]
dt_res = {
'image_id': image_id,
'category_id': category_id,
'bbox': bbox,
'score': score
}
dts_res.append(dt_res)
return dts_res
def test():
dts_res = []
for batch_id, data in enumerate(test_reader()):
nmsed_out_v = exe.run(fluid.default_main_program(),
feed=feeder.feed(data),
fetch_list=[nmsed_out],
return_numpy=False)
if batch_id % 20 == 0:
print("Batch {0}".format(batch_id))
dts_res += get_dt_res(nmsed_out_v, data)
with open("detection_result.json", 'w') as outfile:
json.dump(dts_res, outfile)
print("start evaluate using coco api")
cocoGt = COCO(os.path.join(data_args.data_dir, test_list))
cocoDt = cocoGt.loadRes("detection_result.json")
cocoEval = COCOeval(cocoGt, cocoDt, "bbox")
cocoEval.evaluate()
cocoEval.accumulate()
cocoEval.summarize()
test()