zeakey / deepskeleton Goto Github PK
View Code? Open in Web Editor NEWCode for CVPR2016 and TIP papers about Object Skeleton Detection with Fully Convolutional Neural Networks.
Home Page: http://kaizhao.net/deepsk
Code for CVPR2016 and TIP papers about Object Skeleton Detection with Fully Convolutional Neural Networks.
Home Page: http://kaizhao.net/deepsk
now,i want to fine-tuning the model ,in my data,how could i make my own images' labels?
i see the example in sk506, the groundtruth of 10074.png 's value is range from 0 to 64.
befor,i have used the FCN ,the label is 0 or 1,but deepskeleton is 0 to 64. so , i want to know, in deepskeleton ,the label is 0 to 255? or oehers?
thanks!
Hi, @zeakey
i want to download model on SK506 dataset, but the link is failed, can you Upload your model again? thank you!
HI, @zeakey
Thanks for your codes and I find it very useful for me. But when I run "result.py" ,I find some problem as follow:
F0316 10:43:13.218327 4140 layer.hpp:324] Not Implemented Yet
*** Check failure stack trace: ***
I've checked that "coord_map()" is defined at "layer.hpp:324", while it hasn't been implemented:"NOT_IMPLEMENTED;";but it has been called in "crop_layer.cpp:58"
Have you met the same question before?And how should I do to solve the problem?
Thanks!
Hi, calling this net = caffe.Net(model_root+'deploy.prototxt','sk506_it14k.caffemodel', caffe.TEST)
leads to this error for me:
F0813 19:57:16.217373 1316 crop_layer.cpp:68] Check failed: bottom[0]->shape(i) - crop_offset >= bottom[1]->shape(i) (2 vs. 3) invalid crop parameters in dimension: 1
*** Check failure stack trace: ***
Any ideas? Here's more context:
I0813 19:57:16.214231 1316 net.cpp:141] Setting up conv5_3
I0813 19:57:16.214290 1316 net.cpp:148] Top shape: 1 512 36 36 (663552)
I0813 19:57:16.214304 1316 net.cpp:156] Memory required for data: 895976576
I0813 19:57:16.214329 1316 layer_factory.hpp:77] Creating layer relu5_3
I0813 19:57:16.214351 1316 net.cpp:91] Creating Layer relu5_3
I0813 19:57:16.214366 1316 net.cpp:425] relu5_3 <- conv5_3
I0813 19:57:16.214390 1316 net.cpp:386] relu5_3 -> conv5_3 (in-place)
I0813 19:57:16.214411 1316 net.cpp:141] Setting up relu5_3
I0813 19:57:16.214427 1316 net.cpp:148] Top shape: 1 512 36 36 (663552)
I0813 19:57:16.214439 1316 net.cpp:156] Memory required for data: 898630784
I0813 19:57:16.214453 1316 layer_factory.hpp:77] Creating layer score-dsn-2
I0813 19:57:16.214473 1316 net.cpp:91] Creating Layer score-dsn-2
I0813 19:57:16.214488 1316 net.cpp:425] score-dsn-2 <- conv2_2_relu2_2_0_split_1
I0813 19:57:16.214516 1316 net.cpp:399] score-dsn-2 -> score-dsn2
I0813 19:57:16.215029 1316 net.cpp:141] Setting up score-dsn-2
I0813 19:57:16.215055 1316 net.cpp:148] Top shape: 1 2 284 284 (161312)
I0813 19:57:16.215068 1316 net.cpp:156] Memory required for data: 899276032
I0813 19:57:16.215088 1316 layer_factory.hpp:77] Creating layer upsample-2
I0813 19:57:16.215111 1316 net.cpp:91] Creating Layer upsample-2
I0813 19:57:16.215126 1316 net.cpp:425] upsample-2 <- score-dsn2
I0813 19:57:16.215144 1316 net.cpp:399] upsample-2 -> score-dsn2-up
I0813 19:57:16.217154 1316 net.cpp:141] Setting up upsample-2
I0813 19:57:16.217214 1316 net.cpp:148] Top shape: 1 2 570 570 (649800)
I0813 19:57:16.217231 1316 net.cpp:156] Memory required for data: 901875232
I0813 19:57:16.217257 1316 layer_factory.hpp:77] Creating layer crop
I0813 19:57:16.217289 1316 net.cpp:91] Creating Layer crop
I0813 19:57:16.217308 1316 net.cpp:425] crop <- score-dsn2-up
I0813 19:57:16.217325 1316 net.cpp:425] crop <- data_input_0_split_1
I0813 19:57:16.217344 1316 net.cpp:399] crop -> upscore-dsn2
F0813 19:57:16.217373 1316 crop_layer.cpp:68] Check failed: bottom[0]->shape(i) - crop_offset >= bottom[1]->shape(i) (2 vs. 3) invalid crop parameters in dimension: 1
*** Check failure stack trace: ***
[I 19:57:19.048 NotebookApp] KernelRestarter: restarting kernel (1/5)
Thanks!
Hi @zeakey ,
Thank you very much for your code to give me a lot of help, but when I train your datasets of sk506, I got initial loss around 10000, is that normal? And second question is when I use the program of demo.py, the caffemodel is sk506_it14k.caffemodel, I got error is "cudasuccess(29 vs. 0) driver shutting down"? Have you ever met before? And I put the question on google, the answer told me to check the BVLC/caffe#2016. How should I do to solve the problem? Any advice would be helpful to me.
Thanks!!!
Samylee
hello, i have a question about how to train this model. you see after deconvolution operation there are 5 score-dsn-up. i want to know whether you firstly train every score-dsn-up to look like the original input and then concat the score-dsn-ups to train the final output to look like the original input. i mean whether there are there two training step or you just have one training step
hello, I find the url of dataset on Object Skeleton SK-LARGE
is invalid. can you check it?
Can your code output 3d coordinates of the skeleton over time when fed with video stream frames?
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.