Hello, Liu:

I am running your code on Ubuntu16.04, however, when I run python demo.py, the following two lines showed repeatedly:

QObject::moveToThread: Current thread (0x55c07be39a60) is not the object's thread (0x55c07c164490). Cannot move to target thread (0x55c07be39a60)

Is it a normal phenomenon or there's something wrong with my operation? Thank you for your time.

                                                                                                                                 Dan Meng

There are some inconsistency of gt labels from CTW-1500-reg-label.zip and ctw1500.zip. For example, in 1017.txt, there are 21 instances in CTW-1500-reg-label.zip and 20 instances in ctw1500.zip.

Hi,

Thank you so much for your repository!

In voc_eval_polygon.py, it seems that you are accumulating tp, fp through all the test images? And then calculating prec, rec at the end. Let me know if I misunderstand it.

But I think, at least in the paper, we should calculate prec and recall for each image, and then taking the average of these precision and recall.

Thanks

It seems that the 14points generated do not follow the clockwise or anti-clockwise order before
feeding into the function cv2.drawcontours in ctw1500_loader.py? could you explain why? Thanks a lot.

I am labeling my own dataset with the Labeling tool you provided, but I don't know the exact steps when labeling curved text. Maybe the guiding gif is too quick and I missed some key points? Waiting for more accurate guidance.

Hello!

The annotations in the CTW1500 retain the text reading order, which is very percious in data annotation. And the results on the CTW1500 can be reproduced very successfully.

I also try to reproduce the results of CTD+TLOC on Total-Text and MSRA-TD500. Then I need to convert these two data sets into CTW's annotation format, which means that their reading order also needs to be retained.

How can I complete such a conversion? In the original paper for Total-Text "Because the annotated numbers for this dataset are not fixed and the point positions are subjectively determined, we extracted the text boundaries and used a similar LSI method to resample the annotated points."

Could you give more details on the conversion process?

In voc_eval_polygon.py line 148 to 163 you have not applied offset as you did from171 to 175. Do you think it is a bug?

 # go down dets and mark TPs and FPs
    nd = len(image_ids)
    tp = np.zeros(nd)
    fp = np.zeros(nd)
    for d in range(nd):
        R = class_recs[image_ids[d]]
        bb = BB[d] # mask rcnn
        det_bbox = bb[:]
        pts = [(det_bbox[j], det_bbox[j+1]) for j in xrange(0,len(bb),2)]
        try:
            pdet = Polygon(pts)
        except Exception as e:
            print(e)
            continue
        if not pdet.is_valid: 
            print('predicted polygon has intersecting sides.')
            # print(pts, image_ids[d])
            continue

I can't evaluate my results without changing the part. I changed it like that

 # go down dets and mark TPs and FPs
    nd = len(image_ids)
    tp = np.zeros(nd)
    fp = np.zeros(nd)
    for d in range(nd):
        R = class_recs[image_ids[d]]
        bb = BB[d] # mask rcnn
        det_bbox = bb[:]
        pts = [(det_bbox[j] + det_bbox[0], det_bbox[j+1] + det_bbox[1]) for j in xrange(0,len(bb),2)]
        pts = pts[2:]
        try:
            pdet = Polygon(pts)
        except Exception as e:
            print(e)
            continue
        if not pdet.is_valid: 
            import ipdb; ipdb.set_trace()
            print('predicted polygon has intersecting sides.')
            # print(pts, image_ids[d])
            continue

I met a problem during compiling caffe, I post the screenshot image as below. The problem is "box_annotator_ohem_layer.cu(50): error: a template argument may not reference a local type"; I add "-std=C++11" in makefile of caffe, however , the problem is still there. So have you met the problem during compiling caffe?

I use Ubuntu16.04, cuda9.0, cudnn7.0.3;

Hi, Thanks for the repo, very useful for a project I'm trying to do. I tried running the demo on my MacBook Pro (no CUDA GPU), and it seems a bit of effort to get Caffe up and running on a Mac. In any case, I work mostly on PyTorch. I can try porting your checkpoint and prototxt to PyTorch, but do you know if anyone has already done this? Thanks

HI

I am working on ubutu17.10 and installed caffe prebuilt bainaries for ubuntu>17.04. Everything was setup and when I run the demo.py in tools, I get this error

WARNING: Logging before InitGoogleLogging() is written to STDERR W0406 18:33:40.810593 4823 _caffe.cpp:139] DEPRECATION WARNING - deprecated use of Python interface W0406 18:33:40.810638 4823 _caffe.cpp:140] Use this instead (with the named "weights" parameter): W0406 18:33:40.810642 4823 _caffe.cpp:142] Net('../models/ctd/test_ctd_tloc.prototxt', 1, weights='../output/ctd_tloc.caffemodel') [libprotobuf ERROR google/protobuf/text_format.cc:298] Error parsing text-format caffe.NetParameter: 3767:25: Message type "caffe.LayerParameter" has no field named "psroi_pooling_param". F0406 18:33:40.812146 4823 upgrade_proto.cpp:88] Check failed: ReadProtoFromTextFile(param_file, param) Failed to parse NetParameter file: ../models/ctd/test_ctd_tloc.prototxt *** Check failure stack trace: *** Aborted (core dumped)

Please help me to resolve the issue.

With Regards

Hi Sir: I had compiled it on windows 7 with and without CUDA 8.0(cudnn 5.1 and cudnn 6 .0), but when run test_net.py, it could not get correct porposal, all scores of proposal is 0.5? And also test on Linux, it is ok with GPU, but with CPU, the scores of proposal is correct and same with GPU when read bottom from cpu_data, but the final result is not detecting the text? So, what make it only work on Linux, how to port to Windows? what is the keypoint? Thank you very much! @yl9393 @Yuliang-Liu BR bygreencn

Hi, thanks for your great work. I downloaded the windows version of annotation tool and annotated my image using it. The result text file contains only positions of 14 points and label. To make my data full format of ctw_1500 dataset, what steps should I follow? Could you instruct me how to get full xml file from the annotation file?

The annotation results contains only label and segs. How can I get box and pts?

Thanks in advance.

Hello, can I submit my results to this list? Does it have to be accompanied by a paper explaining the method? I look forward to your answer very much.