def im_proposals(model, im):
"""Generate RPN proposals on a single image."""
inputs = {}
inputs['data'], inputs['im_info'] = _get_image_blob(im)
for k, v in inputs.items():
workspace.FeedBlob(core.ScopedName(k), v.astype(np.float32, copy=False))
workspace.RunNet(model.net.Proto().name)
scale = inputs['im_info'][0, 2]
if cfg.FPN.FPN_ON and cfg.FPN.MULTILEVEL_RPN:
k_max = cfg.FPN.RPN_MAX_LEVEL
k_min = cfg.FPN.RPN_MIN_LEVEL
rois_names = [
core.ScopedName('rpn_rois_fpn' + str(l))
for l in range(k_min, k_max + 1)]
score_names = [
core.ScopedName('rpn_roi_probs_fpn' + str(l))
for l in range(k_min, k_max + 1)]
blobs = workspace.FetchBlobs(rois_names + score_names)
# Combine predictions across all levels and retain the top scoring
boxes = np.concatenate(blobs[:len(rois_names)])
scores = np.concatenate(blobs[len(rois_names):]).squeeze()
# TODO(rbg): NMS again?
inds = np.argsort(-scores)[:cfg.TEST.RPN_POST_NMS_TOP_N]
scores = scores[inds]
boxes = boxes[inds, :]
else:
boxes, scores = workspace.FetchBlobs(
[core.ScopedName('rpn_rois'), core.ScopedName('rpn_roi_probs')])
scores = scores.squeeze()
# Column 0 is the batch index in the (batch ind, x1, y1, x2, y2) encoding,
# so we remove it since we just want to return boxes
boxes = boxes[:, 1:] / scale
return boxes, scores
def test_nativeop_forward(
model,
inputs: List[Tuple[str, np.ndarray]],
reference_outputs: List[Tuple[str, Optional[np.ndarray]]],
metrics: List[TestMetric] = DefaultOpMetrics()
) -> List[Any]:
""" Tests a framework-native operator's forward method on list of given
input tensors, without surrounding overhead.
Similar to `test_op_forward`, but does not incur the overhead of
converting from/to numpy arrays, creating multiple sessions etc.
@param op An operator to test.
@param inputs An array of inputs to pass to the operator.
@param reference_outputs An array of reference outputs.
@param metrics A list of TestMetric objects to measure.
@return List of test metric results for every output.
"""
assert isinstance(inputs, (list, tuple)) == True
normal_metrics = [m for m in metrics if m.reruns == 0]
rerun_metrics = [m for m in metrics if m.reruns > 0]
for (name, inp) in inputs:
workspace.FeedBlob(name, inp)
num_outputs = len(reference_outputs)
# Create a single session
workspace.CreateNet(model.net)
for metric in normal_metrics:
metric.begin(inputs)
workspace.RunNet(model.net)
outputs = workspace.FetchBlobs([name for (name, out) in reference_outputs])
for metric in normal_metrics:
metric.end(outputs)
for metric in rerun_metrics:
for i in range(metric.reruns):
metric.begin(inputs)
workspace.RunNet(model.net)
outputs = workspace.FetchBlobs([name for (name, out) in reference_outputs])
metric.end(outputs)
if not isinstance(outputs, (list, tuple)):
outputs = [outputs]
results = []
for i in range(num_outputs):
# Execute metrics.
for metric in metrics:
result = metric.measure(inputs, outputs[i], (reference_outputs[i])[1])
results.append(result)
summary = metric.measure_summary(inputs, outputs[i],
reference_outputs[i][1])
print("{} on native inference for output {}: {}".format(
type(metric).__name__, i, summary))
return results
def _test_std(self):
root_dir = osp.join('/private', 'home', 'xinleic', 'pyramid')
cfg_file = osp.join(root_dir, 'configs', 'visual_genome', 'e2e_faster_rcnn_R-50-FPN_1x.yaml')
merge_cfg_from_file(cfg_file)
cfg.NUM_GPUS = 1
cfg.TEST.RPN_PRE_NMS_TOP_N = 100
cfg.TEST.RPN_POST_NMS_TOP_N = 20
assert_and_infer_cfg()
test_weight = osp.join(root_dir, 'outputs', 'train', 'visual_genome_train',
'e2e_faster_rcnn_R-50-FPN_1x', 'RNG_SEED#3', 'model_final.pkl')
model = test_engine.initialize_model_from_cfg(test_weight, gpu_id=0)
dataset = JsonDataset('visual_genome_val')
roidb = dataset.get_roidb()
num_images = len(roidb)
num_classes = cfg.MODEL.NUM_CLASSES
entry = roidb[1]
im = cv2.imread(entry['image'])
max_level = cfg.FPN.RPN_MAX_LEVEL
min_level = cfg.FPN.RPN_MIN_LEVEL
# input: rpn_cls_probs_fpn2, rpn_bbox_pred_fpn2
# output: rpn_rois_fpn2, rpn_roi_probs_fpn2
with utils.c2.NamedCudaScope(0):
# let's manually do the testing here
inputs, im_scale = _get_blobs(im, None, cfg.TEST.SCALE, cfg.TEST.MAX_SIZE)
for k, v in inputs.items():
workspace.FeedBlob(core.ScopedName(k), v)
workspace.RunNet(model.net.Proto().name)
cls_probs = [core.ScopedName('rpn_cls_probs_fpn%d' % i) for i in range(min_level, max_level+1)]
box_preds = [core.ScopedName('rpn_bbox_pred_fpn%d' % i) for i in range(min_level, max_level+1)]
rpn_rois = [core.ScopedName('rpn_rois_fpn%d' % i) for i in range(min_level, max_level+1)]
rpn_roi_probs = [core.ScopedName('rpn_roi_probs_fpn%d' % i) for i in range(min_level, max_level+1)]
cls_probs = workspace.FetchBlobs(cls_probs)
box_preds = workspace.FetchBlobs(box_preds)
rpn_rois = workspace.FetchBlobs(rpn_rois)
rpn_roi_probs = workspace.FetchBlobs(rpn_roi_probs)
rpn_rois = np.vstack(rpn_rois)
rpn_roi_probs = np.vstack(rpn_roi_probs)
# remove the image dimension
rpn_rois = rpn_rois[:, 1:]
boxes = np.hstack([rpn_rois, rpn_roi_probs])
im_name = osp.splitext(osp.basename(entry['image']))[0]
utils.vis.vis_one_image(im[:, :, ::-1],
'{:s}-std-output'.format(im_name),
osp.join(root_dir, 'tests'),
boxes,
segms=None,
keypoints=None,
thresh=0.,
box_alpha=0.8,
dataset=dataset,
show_class=False)
workspace.ResetWorkspace()
im_info = inputs['im_info'].astype(np.float32)
return cls_probs, box_preds, im_info, im, im_name, root_dir, dataset
def test_log_barrier(self, X):
param = core.BlobReference("X")
workspace.FeedBlob(param, X)
train_init_net, train_net = self.get_training_nets()
reg = regularizer.LogBarrier(1.0)
output = reg(train_net,
train_init_net,
param,
by=RegularizationBy.ON_LOSS)
reg(
train_net,
train_init_net,
param,
grad=None,
by=RegularizationBy.AFTER_OPTIMIZER,
)
workspace.RunNetOnce(train_init_net)
workspace.RunNetOnce(train_net)
def ref(X):
return (
np.array(np.sum(-np.log(np.clip(X, 1e-9, None))) * 0.5).astype(
np.float32),
np.clip(X, 1e-9, None),
)
for x, y in zip(workspace.FetchBlobs([output, param]), ref(X)):
npt.assert_allclose(x, y, rtol=1e-3)
def testFetchBlobs(self):
s1 = "test1"
s2 = "test2"
workspace.FeedBlob('s1', s1)
workspace.FeedBlob('s2', s2)
fetch1, fetch2 = workspace.FetchBlobs(['s1', 's2'])
self.assertEquals(s1, fetch1)
self.assertEquals(s2, fetch2)
def testFetchBlobs(self):
s1 = b"test1"
s2 = b"test2"
workspace.FeedBlob("s1", s1)
workspace.FeedBlob("s2", s2)
fetch1, fetch2 = workspace.FetchBlobs(["s1", "s2"])
self.assertEquals(s1, fetch1)
self.assertEquals(s2, fetch2)
def test_grad(self, size):
X = np.random.random_sample(size)
workspace.ResetWorkspace()
workspace.FeedBlob("X", X.astype(np.float32))
net = core.Net("negate_grad_test")
Y = net.NegateGradient(["X"], ["Y"])
grad_map = net.AddGradientOperators([Y])
workspace.RunNetOnce(net)
# check X_grad == negate of Y_grad
x_val, y_val = workspace.FetchBlobs(['X', 'Y'])
x_grad_val, y_grad_val = workspace.FetchBlobs(
[grad_map['X'], grad_map['Y']])
np.testing.assert_array_equal(x_val, y_val)
np.testing.assert_array_equal(x_grad_val, y_grad_val * (-1))
def test_transpose_network(self, batch_size, channels, height, width, seed,
kernel):
net = core.Net("net")
net.Conv(["X", "w1", "b1"], ["c1"], stride=1, pad=0, kernel=kernel)
net.Conv(["X", "w2", "b2"], ["c2"], stride=1, pad=0, kernel=kernel)
# c1 and c2: batch_size, 2*channels, height - kernel + 1, width - kernel + 1
net.Conv(["c1", "w3", "b3"], ["c3"], stride=1, pad=0, kernel=kernel)
net.Conv(["c1", "w4", "b4"], ["c4"], stride=1, pad=0, kernel=kernel)
# c3 and c4: batch_size, 2*channels, height - 2*kernel + 2, width - 2*kernel + 2
net.Flatten(["c3"], "c3f")
net.Flatten(["c4"], "c4f")
net.Flatten(["X"], "Xf")
net.Concat(["c3f", "c4f", "Xf"], ["out", "split_info"],
axis=1,
add_axis=0)
np.random.seed(seed)
workspace.ResetWorkspace()
tu.randBlobFloat32("X", batch_size, channels, height, width)
tu.randBlobsFloat32(["w1", "w2"], 2 * channels, channels, kernel,
kernel)
tu.randBlobsFloat32(["b1", "b2"], 2 * channels)
tu.randBlobsFloat32(["w3", "w4"], 4 * channels, 2 * channels, kernel,
kernel)
tu.randBlobsFloat32(["b3", "b4"], 4 * channels)
all_inp_names = ["X", "w1", "w2", "b1", "b2", "w3", "w4", "b3", "b4"]
all_input = workspace.FetchBlobs(all_inp_names)
workspace.RunNetOnce(net)
preTransformC1 = workspace.FetchBlob("c1")
preTransformC3 = workspace.FetchBlob("c3")
preTransformOut = workspace.FetchBlob("out")
nn = ng.NNModule(net)
preTransformNumOperators = len(nn.operators)
preTransformNumTensors = len(nn.tensors)
transpose_network(nn)
new_netdef = nn.convertToCaffe2Proto()
postTransformNumOperators = len(nn.operators)
postTransformNumTensors = len(nn.tensors)
# The minimal number of additional operators and tensors is at least one
# NCHW2NHWC operator and tensor for each channel-based input tensor
# and a NHWC2NCHW operator and tensor for the output of each convolution
# X, w1, w2, w3, w4 are channel-based inputs
# c1, c2, c3, c4 are the outputs of convolutions
# i.e. a total of 9.
self.assertEqual(postTransformNumOperators,
preTransformNumOperators + 9,
"expected 9 additional operators")
self.assertEqual(postTransformNumTensors, preTransformNumTensors + 9,
"expected 9 additional tensors")
workspace.ResetWorkspace()
for name, val in zip(all_inp_names, all_input):
workspace.FeedBlob(name, val)
workspace.RunNetOnce(new_netdef)
postTransformC1 = workspace.FetchBlob("c1")
postTransformC3 = workspace.FetchBlob("c3")
postTransformOut = workspace.FetchBlob("out")
np.testing.assert_almost_equal(postTransformC1, preTransformC1, 1)
np.testing.assert_almost_equal(postTransformC3, preTransformC3, 1)
np.testing.assert_almost_equal(postTransformOut, preTransformOut, 1)
def im_proposals(model, im):
"""Generate RPN proposals on a single image."""
inputs = {}
inputs['data'], inputs['im_info'] = _get_image_blob(im)
scale = inputs['im_info'][0, 2]
for k, v in inputs.items():
workspace.FeedBlob(core.ScopedName(k), v.astype(np.float32,
copy=False))
workspace.RunNet(model.net.Proto().name)
if cfg.FPN.FPN_ON and cfg.FPN.MULTILEVEL_RPN:
k_max = cfg.FPN.RPN_MAX_LEVEL
k_min = cfg.FPN.RPN_MIN_LEVEL
rois_names = [
core.ScopedName('rpn_rois_fpn' + str(l))
for l in range(k_min, k_max + 1)
]
score_names = [
core.ScopedName('rpn_roi_probs_fpn' + str(l))
for l in range(k_min, k_max + 1)
]
blobs = workspace.FetchBlobs(rois_names + score_names)
# Combine predictions across all levels and retain the top scoring
boxes = np.concatenate(blobs[:len(rois_names)])
scores = np.concatenate(blobs[len(rois_names):]).squeeze()
# Discussion: one could do NMS again after combining predictions from
# the different FPN levels. Conceptually, it's probably the right thing
# to do. For arbitrary reasons, the original FPN RPN implementation did
# not do another round of NMS.
inds = np.argsort(-scores)[:cfg.TEST.RPN_POST_NMS_TOP_N]
scores = scores[inds]
boxes = boxes[inds, :]
else:
boxes, scores = workspace.FetchBlobs(
[core.ScopedName('rpn_rois'),
core.ScopedName('rpn_roi_probs')])
scores = scores.squeeze()
# Column 0 is the batch index in the (batch ind, x1, y1, x2, y2) encoding,
# so we remove it since we just want to return boxes
# Scale proposals back to the original input image scale
boxes = boxes[:, 1:] / scale
return boxes, scores
def _run_general_op_cpu(self, op_name, blobs_in, values_in, blobs_out, **kwargs):
with core.DeviceScope(core.DeviceOption(caffe2_pb2.CPU, 0)):
op = core.CreateOperator(op_name, blobs_in, blobs_out, **kwargs)
for name, value in zip(blobs_in, values_in):
workspace.FeedBlob(name, value)
workspace.RunOperatorOnce(op)
values_out = workspace.FetchBlobs(blobs_out)
workspace.ResetWorkspace()
return values_out
def im_detect_bbox(model, im, timers=None):
"""Generate RetinaNet detections on a single image."""
if timers is None:
timers = defaultdict(Timer)
# Although anchors are input independent and could be precomputed,
# recomputing them per image only brings a small overhead
anchors = _create_cell_anchors()
timers['im_detect_bbox'].tic()
k_max, k_min = cfg.FPN.RPN_MAX_LEVEL, cfg.FPN.RPN_MIN_LEVEL
A = cfg.RETINANET.SCALES_PER_OCTAVE * len(cfg.RETINANET.ASPECT_RATIOS)
inputs = {}
inputs['data'], inputs['im_info'] = _get_image_blob(im)
cls_probs, box_preds = [], []
for lvl in range(k_min, k_max + 1):
suffix = 'fpn{}'.format(lvl)
cls_probs.append(core.ScopedName('retnet_cls_prob_{}'.format(suffix)))
box_preds.append(core.ScopedName('retnet_bbox_pred_{}'.format(suffix)))
for k, v in inputs.items():
workspace.FeedBlob(core.ScopedName(k), v.astype(np.float32,
copy=False))
workspace.RunNet(model.net.Proto().name)
scale = inputs['im_info'][0, 2]
cls_probs = workspace.FetchBlobs(cls_probs)
box_preds = workspace.FetchBlobs(box_preds)
# here the boxes_all are [x0, y0, x1, y1, score]
boxes_all = defaultdict(list)
cnt = 0
for lvl in range(k_min, k_max + 1):
# create cell anchors array
stride = 2.**lvl
cell_anchors = anchors[lvl]
# fetch per level probability
cls_prob = cls_probs[cnt]
box_pred = box_preds[cnt]
cls_prob = cls_prob.reshape(
(cls_prob.shape[0], A, int(cls_prob.shape[1] / A),
cls_prob.shape[2], cls_prob.shape[3]))
box_pred = box_pred.reshape(
(box_pred.shape[0], A, 4, box_pred.shape[2], box_pred.shape[3]))
cnt += 1
if cfg.RETINANET.SOFTMAX:
cls_prob = cls_prob[:, :, 1::, :, :]
cls_prob_ravel = cls_prob.ravel()
# In some cases [especially for very small img sizes], it's possible that
# candidate_ind is empty if we impose threshold 0.05 at all levels. This
# will lead to errors since no detections are found for this image. Hence,
# for lvl 7 which has small spatial resolution, we take the threshold 0.0
th = cfg.RETINANET.INFERENCE_TH if lvl < k_max else 0.0
candidate_inds = np.where(cls_prob_ravel > th)[0]
if (len(candidate_inds) == 0):
continue
pre_nms_topn = min(cfg.RETINANET.PRE_NMS_TOP_N, len(candidate_inds))
inds = np.argpartition(cls_prob_ravel[candidate_inds],
-pre_nms_topn)[-pre_nms_topn:]
inds = candidate_inds[inds]
inds_5d = np.array(np.unravel_index(inds, cls_prob.shape)).transpose()
classes = inds_5d[:, 2]
anchor_ids, y, x = inds_5d[:, 1], inds_5d[:, 3], inds_5d[:, 4]
scores = cls_prob[:, anchor_ids, classes, y, x]
boxes = np.column_stack((x, y, x, y)).astype(dtype=np.float32)
boxes *= stride
boxes += cell_anchors[anchor_ids, :]
if not cfg.RETINANET.CLASS_SPECIFIC_BBOX:
box_deltas = box_pred[0, anchor_ids, :, y, x]
else:
box_cls_inds = classes * 4
box_deltas = np.vstack([
box_pred[0, ind:ind + 4, yi, xi]
for ind, yi, xi in zip(box_cls_inds, y, x)
])
pred_boxes = (box_utils.bbox_transform(boxes, box_deltas)
if cfg.TEST.BBOX_REG else boxes)
pred_boxes /= scale
pred_boxes = box_utils.clip_tiled_boxes(pred_boxes, im.shape)
box_scores = np.zeros((pred_boxes.shape[0], 5))
box_scores[:, 0:4] = pred_boxes
box_scores[:, 4] = scores
for cls in range(1, cfg.MODEL.NUM_CLASSES):
inds = np.where(classes == cls - 1)[0]
if len(inds) > 0:
boxes_all[cls].extend(box_scores[inds, :])
timers['im_detect_bbox'].toc()
# Combine predictions across all levels and retain the top scoring by class
timers['misc_bbox'].tic()
detections = []
for cls, boxes in boxes_all.items():
cls_dets = np.vstack(boxes).astype(dtype=np.float32)
# do class specific nms here
keep = box_utils.nms(cls_dets, cfg.TEST.NMS)
cls_dets = cls_dets[keep, :]
out = np.zeros((len(keep), 6))
out[:, 0:5] = cls_dets
out[:, 5].fill(cls)
detections.append(out)
# detections (N, 6) format:
# detections[:, :4] - boxes
# detections[:, 4] - scores
# detections[:, 5] - classes
detections = np.vstack(detections)
# sort all again
inds = np.argsort(-detections[:, 4])
detections = detections[inds[0:cfg.TEST.DETECTIONS_PER_IM], :]
# Convert the detections to image cls_ format (see core/test_engine.py)
num_classes = cfg.MODEL.NUM_CLASSES
cls_boxes = [[] for _ in range(cfg.MODEL.NUM_CLASSES)]
for c in range(1, num_classes):
inds = np.where(detections[:, 5] == c)[0]
cls_boxes[c] = detections[inds, :5]
timers['misc_bbox'].toc()
return cls_boxes
def make_nps(xs):
assert isinstance(xs,
list), 'ERROR: should pass list of names of the blobs'
return workspace.FetchBlobs(xs)
def test_record_queue(self):
num_prod = 8
num_consume = 3
schema = Struct(
('floats', Map(Scalar(np.int32), Scalar(np.float32))), )
contents_raw = [
[1, 2, 3], # len
[11, 21, 22, 31, 32, 33], # key
[1.1, 2.1, 2.2, 3.1, 3.2, 3.3], # value
]
contents = from_blob_list(schema, contents_raw)
ds = Dataset(schema)
net = core.Net('init')
ds.init_empty(net)
content_blobs = NewRecord(net, contents)
FeedRecord(content_blobs, contents)
writer = ds.writer(init_net=net)
writer.write_record(net, content_blobs)
reader = ds.reader(init_net=net)
# prepare receiving dataset
rec_dataset = Dataset(contents, name='rec')
rec_dataset.init_empty(init_net=net)
rec_dataset_writer = rec_dataset.writer(init_net=net)
workspace.RunNetOnce(net)
queue = RecordQueue(contents, num_threads=num_prod)
def process(net, fields):
new_fields = []
for f in fields.field_blobs():
new_f = net.Copy(f)
new_fields.append(new_f)
new_fields = from_blob_list(fields, new_fields)
return new_fields
q_reader, q_step, q_exit, fields = queue.build(reader, process)
producer_step = core.execution_step('producer', [q_step, q_exit])
consumer_steps = []
for i in range(num_consume):
name = 'queue_reader_' + str(i)
net_consume = core.Net(name)
should_stop, fields = q_reader.read_record(net_consume)
step_consume = core.execution_step(name, net_consume)
name = 'dataset_writer_' + str(i)
net_dataset = core.Net(name)
rec_dataset_writer.write(net_dataset, fields.field_blobs())
step_dataset = core.execution_step(name, net_dataset)
step = core.execution_step('consumer_' + str(i),
[step_consume, step_dataset],
should_stop_blob=should_stop)
consumer_steps.append(step)
consumer_step = core.execution_step('consumers',
consumer_steps,
concurrent_substeps=True)
work_steps = core.execution_step('work',
[producer_step, consumer_step],
concurrent_substeps=True)
plan = core.Plan('test')
plan.AddStep(work_steps)
core.workspace.RunPlan(plan)
data = workspace.FetchBlobs(rec_dataset.get_blobs())
self.assertEqual(6, sum(data[0]))
self.assertEqual(150, sum(data[1]))
self.assertAlmostEqual(15, sum(data[2]), places=5)
def im_detect_bbox(model, im, timers=None, model1=None):
"""Generate RetinaNet detections on a single image."""
if timers is None:
timers = defaultdict(Timer)
if model1 is None and os.environ.get('COSIM'):
print("cosim must has model1")
fp32_ws_name = "__fp32_ws__"
int8_ws_name = "__int8_ws__"
# Although anchors are input independent and could be precomputed,
# recomputing them per image only brings a small overhead
anchors = _create_cell_anchors()
timers['im_detect_bbox'].tic()
timers['data1'].tic()
k_max, k_min = cfg.FPN.RPN_MAX_LEVEL, cfg.FPN.RPN_MIN_LEVEL
A = cfg.RETINANET.SCALES_PER_OCTAVE * len(cfg.RETINANET.ASPECT_RATIOS)
inputs = {}
inputs['data'], im_scale, inputs['im_info'] = \
blob_utils.get_image_blob(im, cfg.TEST.SCALE, cfg.TEST.MAX_SIZE, cfg.TEST.SIZEFIX)
cls_probs, box_preds = [], []
for lvl in range(k_min, k_max + 1):
suffix = 'fpn{}'.format(lvl)
cls_probs.append(core.ScopedName('retnet_cls_prob_{}'.format(suffix)))
box_preds.append(core.ScopedName('retnet_bbox_pred_{}'.format(suffix)))
for k, v in inputs.items():
if os.environ.get('COSIM'):
workspace.SwitchWorkspace(int8_ws_name, True)
workspace.FeedBlob(core.ScopedName(k), v.astype(np.float32,
copy=False))
if os.environ.get('COSIM'):
workspace.SwitchWorkspace(fp32_ws_name, True)
workspace.FeedBlob(core.ScopedName(k),
v.astype(np.float32, copy=False))
timers['data1'].toc()
if os.environ.get('EPOCH2OLD') == "1":
workspace.RunNet(model.net.Proto().name)
timers['run'].tic()
if os.environ.get('INT8INFO') == "1":
algorithm = AbsmaxCalib()
kind = os.environ.get('INT8CALIB')
if kind == "moving_average":
ema_alpha = 0.5
algorithm = EMACalib(ema_alpha)
elif kind == "kl_divergence":
kl_iter_num_for_range = os.environ.get('INT8KLNUM')
if not kl_iter_num_for_range:
kl_iter_num_for_range = 100
kl_iter_num_for_range = int(kl_iter_num_for_range)
algorithm = KLCalib(kl_iter_num_for_range)
calib = Calibrator(algorithm)
calib.RunCalibIter(workspace, model.net.Proto())
else:
if os.environ.get('COSIM'):
with open("int8.txt", "wb") as p:
p.write(str(model.net.Proto()))
with open("fp32.txt", "wb") as p:
p.write(str(model1.net.Proto()))
for i in range(len(model.net.Proto().op)):
workspace.SwitchWorkspace(int8_ws_name)
int8_inputs = []
for inp in model.net.Proto().op[i].input:
int8_inputs.append(workspace.FetchBlob(str(inp)))
logging.warning(" opint8[{0}] is {1}".format(
i,
model.net.Proto().op[i]))
workspace.RunOperatorOnce(model.net.Proto().op[i])
int8_results = []
for res in model.net.Proto().op[i].output:
int8_results.append(workspace.FetchBlob(str(res)))
workspace.SwitchWorkspace(fp32_ws_name)
fp32_inputs = []
for inp1 in model1.net.Proto().op[i].input:
fp32_inputs.append(workspace.FetchBlob(str(inp1)))
logging.warning(" opfp32[{0}] is {1}".format(
i,
model1.net.Proto().op[i]))
workspace.RunOperatorOnce(model1.net.Proto().op[i])
fp32_results = []
for res1 in model1.net.Proto().op[i].output:
fp32_results.append(workspace.FetchBlob(str(res1)))
if len(int8_inputs) != len(fp32_inputs):
logging.error("Wrong number of inputs")
return
if len(int8_results) != len(fp32_results):
logging.error("Wrong number of outputs")
return
logging.warning("begin to check op[{}] {} input".format(
i,
model.net.Proto().op[i].type))
for k in range(len(int8_inputs)):
if model.net.Proto().op[i].input[k][0] == '_':
continue
#assert_allclose(int8_inputs[k], fp32_inputs[k], **tol)
logging.warning("pass checking op[{0}] {1} input".format(
i,
model.net.Proto().op[i].type))
logging.warning("begin to check op[{0}] {1} output".format(
i,
model.net.Proto().op[i].type))
for j, int8_result in enumerate(int8_results):
if model.net.Proto().op[i].output[j][0] == '_':
continue
#logging.warning("int8_outputis {} and fp32 output is {} ".format(int8_results[j], fp32_results[j]))
#if not compare_utils.assert_allclose(int8_results[j], fp32_results[j], **tol):
if not compare_utils.assert_compare(
int8_result, fp32_results[j], 1e-01,
os.environ.get('COSIM')):
for k, int8_input in enumerate(int8_inputs):
logging.warning("int8_input[{}] is {}".format(
k, int8_input))
logging.warning("fp32_input[{}] is {}".format(
k, fp32_inputs[k]))
logging.warning("pass checking op[{0}] {1} output".format(
i,
model.net.Proto().op[i].type))
else:
workspace.RunNet(model.net.Proto().name)
timers['run'].toc()
cls_probs = workspace.FetchBlobs(cls_probs)
box_preds = workspace.FetchBlobs(box_preds)
# here the boxes_all are [x0, y0, x1, y1, score]
boxes_all = defaultdict(list)
batch_size = cls_probs[0].shape[0]
boxes_all_list = [boxes_all] * batch_size
cnt = 0
for lvl in range(k_min, k_max + 1):
# create cell anchors array
stride = 2.**lvl
cell_anchors = anchors[lvl]
# fetch per level probability
cls_prob = cls_probs[cnt]
box_pred = box_preds[cnt]
cls_prob = cls_prob.reshape(
(cls_prob.shape[0], A, int(cls_prob.shape[1] / A),
cls_prob.shape[2], cls_prob.shape[3]))
box_pred = box_pred.reshape(
(box_pred.shape[0], A, 4, box_pred.shape[2], box_pred.shape[3]))
cnt += 1
if cfg.RETINANET.SOFTMAX:
cls_prob = cls_prob[:, :, 1::, :, :]
for i in range(batch_size):
cls_prob_ravel = cls_prob[i, :].ravel()
# In some cases [especially for very small img sizes], it's possible that
# candidate_ind is empty if we impose threshold 0.05 at all levels. This
# will lead to errors since no detections are found for this image. Hence,
# for lvl 7 which has small spatial resolution, we take the threshold 0.0
th = cfg.RETINANET.INFERENCE_TH if lvl < k_max else 0.0
candidate_inds = np.where(cls_prob_ravel > th)[0]
if (len(candidate_inds) == 0):
continue
pre_nms_topn = min(cfg.RETINANET.PRE_NMS_TOP_N,
len(candidate_inds))
inds = np.argpartition(cls_prob_ravel[candidate_inds],
-pre_nms_topn)[-pre_nms_topn:]
inds = candidate_inds[inds]
inds_4d = np.array(np.unravel_index(
inds, (cls_prob[i, :]).shape)).transpose()
classes = inds_4d[:, 1]
anchor_ids, y, x = inds_4d[:, 0], inds_4d[:, 2], inds_4d[:, 3]
scores = cls_prob[i, anchor_ids, classes, y, x]
boxes = np.column_stack((x, y, x, y)).astype(dtype=np.float32)
boxes *= stride
boxes += cell_anchors[anchor_ids, :]
if not cfg.RETINANET.CLASS_SPECIFIC_BBOX:
box_deltas = box_pred[i, anchor_ids, :, y, x]
else:
box_cls_inds = classes * 4
box_deltas = np.vstack([
box_pred[i, ind:ind + 4, yi, xi]
for ind, yi, xi in zip(box_cls_inds, y, x)
])
pred_boxes = (box_utils.bbox_transform(boxes, box_deltas)
if cfg.TEST.BBOX_REG else boxes)
pred_boxes /= im_scale
pred_boxes = box_utils.clip_tiled_boxes(pred_boxes, im[0].shape)
box_scores = np.zeros((pred_boxes.shape[0], 5))
box_scores[:, 0:4] = pred_boxes
box_scores[:, 4] = scores
for cls in range(1, cfg.MODEL.NUM_CLASSES):
inds = np.where(classes == cls - 1)[0]
if len(inds) > 0:
boxes_all_list[i][cls].extend(box_scores[inds, :])
timers['im_detect_bbox'].toc()
cls_boxes_list = []
for i in range(batch_size):
boxes_all = boxes_all_list[i]
# Combine predictions across all levels and retain the top scoring by class
timers['misc_bbox'].tic()
detections = []
for cls, boxes in boxes_all.items():
cls_dets = np.vstack(boxes).astype(dtype=np.float32)
# do class specific nms here
keep = box_utils.nms(cls_dets, cfg.TEST.NMS)
cls_dets = cls_dets[keep, :]
out = np.zeros((len(keep), 6))
out[:, 0:5] = cls_dets
out[:, 5].fill(cls)
detections.append(out)
# detections (N, 6) format:
# detections[:, :4] - boxes
# detections[:, 4] - scores
# detections[:, 5] - classes
detections = np.vstack(detections)
# sort all again
inds = np.argsort(-detections[:, 4])
detections = detections[inds[0:cfg.TEST.DETECTIONS_PER_IM], :]
# Convert the detections to image cls_ format (see core/test_engine.py)
num_classes = cfg.MODEL.NUM_CLASSES
cls_boxes = [[] for _ in range(cfg.MODEL.NUM_CLASSES)]
for c in range(1, num_classes):
inds = np.where(detections[:, 5] == c)[0]
cls_boxes[c] = detections[inds, :5]
cls_boxes_list.append(cls_boxes)
timers['misc_bbox'].toc()
return cls_boxes_list