light_head_rcnn
light_head_rcnn copied to clipboard
zengarden
No OpKernel was registered to support Op 'NMS' with these attrs
Hello ,
I'm try to run your test.py on my env . but meet the problem below , did I make some mistake ?
(I could inport /lib/lib_kernel/lib_fast_nms/nms_op.py but could't use it .)
Caused by op 'resnet_v1_101_5/NMS', defined at:
File "test.py", line 242, in <module>
eval_all(args)
File "test.py", line 162, in eval_all
proc.start()
File "/usr/lib/python3.4/multiprocessing/process.py", line 105, in start
self._popen = self._Popen(self)
File "/usr/lib/python3.4/multiprocessing/context.py", line 212, in _Popen
return _default_context.get_context().Process._Popen(process_obj)
File "/usr/lib/python3.4/multiprocessing/context.py", line 267, in _Popen
return Popen(process_obj)
File "/usr/lib/python3.4/multiprocessing/popen_fork.py", line 21, in __init__
self._launch(process_obj)
File "/usr/lib/python3.4/multiprocessing/popen_fork.py", line 77, in _launch
code = process_obj._bootstrap()
File "/usr/lib/python3.4/multiprocessing/process.py", line 254, in _bootstrap
self.run()
File "/usr/lib/python3.4/multiprocessing/process.py", line 93, in run
self._target(*self._args, **self._kwargs)
File "test.py", line 107, in worker
func, inputs = load_model(model_file, dev)
File "test.py", line 38, in load_model
net.inference('TEST', inputs)
File "/home/aipr/dennis_codebase/light_head_rcnn/experiments/lizeming/light_head_rcnn.ori_res101.coco.ps_roialign/network_desp.py", line 164, in inference
anchors, num_anchors, is_tfchannel=True, is_tfnms=False)
File "/home/aipr/dennis_codebase/light_head_rcnn/lib/detection_opr/rpn_batched/proposal_opr.py", line 95, in proposal_opr
cur_proposals, nms_thresh, post_nms_topN)
File "<string>", line 43, in nms
File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/framework/op_def_library.py", line 328, in apply_op
op_type_name, name, **keywords)
File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper
op_def=op_def)
File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/framework/ops.py", line 3160, in create_op
op_def=op_def)
File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/framework/ops.py", line 1625, in __init__
self._traceback = self._graph._extract_stack() # pylint: disable=protected-access
InvalidArgumentError (see above for traceback): No OpKernel was registered to support Op 'NMS' with these attrs. Registered devices: [CPU], Registered kernels:
device='GPU'; T in [DT_FLOAT]
[[Node: resnet_v1_101_5/NMS = NMS[T=DT_FLOAT, max_out=1000, nms_overlap_thresh=0.7](resnet_v1_101_5/Gather)]]
@liumusicforever Pls compile nms first. Ensure .so is generated in /lib/lib_kernel/lib_fast_nms/. If not, just run make.sh.
@zengarden , I already ran make.sh and have a fast_nms.so , why still print the error above ?was the path linking problem ?
updata :
I can pass when running python3 nms_op.py , but meet same error message on running python3 nms_test.py
nms_out = nms_op.nms(rois, 0.5, 200)
File "<string>", line 43, in nms
File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/framework/op_def_library.py", line 328, in apply_op
op_type_name, name, **keywords)
File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/framework/op_def_library.py", line 787, in _apply_op_helper
op_def=op_def)
File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/framework/ops.py", line 3160, in create_op
op_def=op_def)
File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/framework/ops.py", line 1625, in __init__
self._traceback = self._graph._extract_stack() # pylint: disable=protected-access
InvalidArgumentError (see above for traceback): No OpKernel was registered to support Op 'NMS' with these attrs. Registered devices: [CPU], Registered kernels:
device='GPU'; T in [DT_FLOAT]
[[Node: NMS = NMS[T=DT_FLOAT, max_out=200, nms_overlap_thresh=0.5](Const)]]
It may have something wrong occur in calling nms_op.nms operasion .
thanks !
NMS works in my machine, if you still have problem, your can change is_tfnms to "True" to directly use tf_nms in network_desp.py.
rois, roi_scores = proposal_opr(
rpn_cls_prob, rpn_bbox_pred, im_info, mode, cfg.stride,
anchors, num_anchors, is_tfchannel=True, is_tfnms=False)
Very thanks for your help ! I found some problem with my process:
- I didn't change
CUDA_VISIBLE_DEVICESvariable to zero (only one gpu) - I can't run this repo with cudnn==7.1.3 , but fix in reinstalling 7.0.4 , here was the error message :
2018-04-24 08:57:14.806152: E tensorflow/stream_executor/cuda/cuda_dnn.cc:378] Loaded runtime CuDNN library: 7103 (compatibility version 7100) but source was compiled with 7004 (compatibility version 7000). If using a binary install, up
grade your CuDNN library to match. If building from sources, make sure the library loaded at runtime matches a compatible version specified during compile configuration.
2018-04-24 08:57:14.806720: F tensorflow/core/kernels/conv_ops.cc:717] Check failed: stream->parent()->GetConvolveAlgorithms( conv_parameters.ShouldIncludeWinogradNonfusedAlgo<T>(), &algorithms)
Aborted (core dumped)
thanks again !
Hello, I got very similar errors. I'm using tensorflow 1.8, cuda-9.0 and cudnn 7.0.5. I tried to change is_tfnms to "True" to directly use tf_nms in network_desp.py but I got different error in stead.
[[Node: resnet_v1_101_6/PSAlignPool = PSAlignPool[T=DT_FLOAT, group_size=7, sample_height=2, sample_width=2, spatial_scale=0.0625](resnet_v1_101_6/Relu, resnet_v1_101_5/concat_3)]]
Caused by op 'resnet_v1_101_6/PSAlignPool', defined at:
File "test.py", line 241, in
InvalidArgumentError (see above for traceback): No OpKernel was registered to support Op 'PSAlignPool' with these attrs. Registered devices: [CPU], Registered kernels: device='GPU'; T in [DT_FLOAT]
[[Node: resnet_v1_101_6/PSAlignPool = PSAlignPool[T=DT_FLOAT, group_size=7, sample_height=2, sample_width=2, spatial_scale=0.0625](resnet_v1_101_6/Relu, resnet_v1_101_5/concat_3)]]
You can check your tensorflow env (is gpu version or not) , the problem may cause that session will create operation on GPU but your enc registered on CPU.
@zengarden hi, your work is very excellent! For my case, i can not use your fast_nms because there is no cuda9 in my machine. if i choose tf_nms, except modifying here:
rois, roi_scores = proposal_opr( rpn_cls_prob, rpn_bbox_pred, im_info, mode, cfg.stride, anchors, num_anchors, is_tfchannel=True, is_tfnms=True)
anywhere else should i modify?
If you only have one GPU,set variable os.environ["CUDA_VISIBLE_DEVICES"] = '0',rather than 1.
Hi, @liumusicforever
When you run test.py, have you met the error as follows?
I really can't solve the problem, please help me, Thx.
@wsycl can you pass nms_test.py in light_head_rcnn/lib/lib_kernel/lib_fast_nms/ ?
and make sure .so in your lib_fast_nms folder
@liumusicforever Thx, I have solved the problems. Can i ask you one more question? How to change the .json file to .odformat?
@wsycl you can change your json metadata like .odgt format , here is the simple example to convert Pascal dataset to odgt (in your case , you can focus on main() function):
import cv2
import os
import xml.etree.cElementTree as ET
import json
def get_img_list(work_path):
image_list = []
for root , subdir , files in os.walk(work_path):
for img_filename in files:
if '.jpg' in img_filename :
image_path = root+'/'+img_filename
annot_filename = img_filename.replace('.jpg','.xml')
annot_path = os.path.join(root,'../','Annotations',annot_filename)
if os.path.exists(image_path) and os.path.exists(annot_path):
image_list.append([image_path,annot_path])
return image_list
def xmlreader(xml_file_path):
boxlist=[]
if not os.path.exists(xml_file_path):
print('No xml file')
return 0,0,boxlist
else:
f=open(xml_file_path).read()
if len(f) < 10:
return 0,0,boxlist
root=ET.fromstring(f)
size = root.find('size')[:]
w = float(size[0].text)
h = float(size[1].text)
objects=root.findall('object')
for ob in objects:
xmin,ymin,xmax,ymax =ob.find('bndbox')[:]
xmin = float(xmin.text)/w
ymin = float(ymin.text)/h
xmax = float(xmax.text)/w
ymax = float(ymax.text)/h
l = [xmin,ymin,xmax,ymax]
boxlist.append(l)
if len(boxlist)==0:
boxlist=[]
return w,h,boxlist
def voc_parser(root,limit = 1000000000):
datas = get_img_list(root)
image_list = []
for i,(img_path,annot_path) in enumerate(datas):
w,h,bboxes = xmlreader(annot_path)
image_list.append([img_path,bboxes])
if i % 1000 == 0:
print ('get training items : {}'.format(len(image_list)))
if i > limit:
return image_list
return image_list
def main():
voc_root = '<path to pascal dataset>'
output = 'out.odgt'
output_list = []
for i ,(img_path , bboxes) in enumerate(voc_parser(voc_root)):
print ('processing {} images : {}'.format(i,img_path))
img = cv2.imread(img_path)
h,w = img.shape[:2]
meta_dict = {
"gtboxes":[],
"fpath":img_path,
"dbName":"COCO",
"dbInfo":{
"vID":"COCO_trainval2014_womini",
"frameID":-1
},
"width":w,
"height":h,
"ID":img_path.split('/')[-1]
}
for bbox in bboxes:
xmin,ymin,xmax,ymax = bbox[:]
xcen = int((xmin + xmax)/2*w)
ycen = int((ymin + ymax)/2*h)
x_w = int((xmax - xmin)/2*w)
y_w = int((ymax - ymin)/2*h)
box_info_dict = {
"box":[
xcen,
ycen,
x_w,
y_w
],
"occ":0,
"tag":"person",
"extra":{
"ignore":0
}
}
meta_dict['gtboxes'].append(box_info_dict)
output_list.append(json.dumps(meta_dict)+ "\n")
with open(output, "w") as f:
f.writelines(output_list)
if __name__ == '__main__':
main()