Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Hog tutorial not working #44

Open
davidbp opened this issue Nov 5, 2017 · 13 comments
Open

Hog tutorial not working #44

davidbp opened this issue Nov 5, 2017 · 13 comments

Comments

@davidbp
Copy link
Contributor

davidbp commented Nov 5, 2017

The tutorial folder here:
http://juliaimages.github.io/ImageFeatures.jl/latest/tutorials/object_detection.html

contains a bash script that is not working since url is not valid. Since there is another url that seems to work maybe you could simply remove the first one.

@kmsquire
Copy link

kmsquire commented Nov 6, 2017

Thanks for the report. Do you feel like submitting a pull request? Since this is a simple change, you could to navigate to https://github.com/JuliaImages/ImageFeatures.jl/blob/master/docs/src/tutorials/object_detection.md and click on the pencil in the upper right to edit the file.

@tejus-gupta
Copy link
Contributor

The tutorial uses this pedestrian dataset - http://cbcl.mit.edu/software-datasets/PedestrianData.html. The download link isn't working right now and I can't find dataset anywhere else.

Thanks for reporting this. I will update the tutorial to use another dataset (by this weekend).

@davidbp
Copy link
Contributor Author

davidbp commented Nov 8, 2017

I have found the data here:

https://github.com/mbeyeler/opencv-machine-learning/tree/78f5f6ecfbbae3f61b378060d60641420331115c/notebooks/data/chapter6

Maybe you could simply add it in the repo and load it from the repo. When compressed its only some MB.

@davidbp
Copy link
Contributor Author

davidbp commented Nov 9, 2017

An even better idea would be to add the data in MLDatasets.jl so other people can use it independently of ImageFeatures easily.

@abhijithch
Copy link

abhijithch commented Nov 22, 2017

Also, there might be a small typo in the get_data.bash file. The name of the Julia file in the bash file(line number 11) is preprocess_data.jl whereas the actual file is preprocess.jl. I could have sent a PR, but this is in the downloaded Tutorial.zip folder.

@tejus-gupta
Copy link
Contributor

@timholy Should I add the contents of Tutorial.zip to this repository itself? Since its on google drive currently, only I can update the content of the tutorial.

@timholy
Copy link
Member

timholy commented Jan 1, 2018

I'm a little reluctant to add a 10MB file to the repository (and once added it becomes a bottleneck on the git history even if we delete it later). What about adding a download script to MLDatasets.jl as suggested by @davidbp? It would be a bit like TestImages.jl.

@Evizero
Copy link
Member

Evizero commented Jan 1, 2018

A simple ad-hoc solution would be to use https://github.com/oxinabox/DataDeps.jl from @oxinabox directly. That package will be the future backend of MLDatasets (see JuliaML/MLDatasets.jl#12), so moving the dataset code to MLDatasets would be possible in the future. The bonus of using DataDeps directly is that it requires a lot less effort to do (tests and documentation) than I am advocating for MLDatasets.

@oxinabox
Copy link

oxinabox commented Jan 4, 2018

I think this is a good use case for it yes.
I've used DataDeps.jl in a couple of quick scripts that had some data depenancy and it worked well and cleanly.
It is still v0.x.y so feedback on a design level would still be appreciated.
I want it to be good for this use case -- when you want the data as a kind of one-off, rather than wanting to go the full level of adding it to MLDatasets.jl etc. (as well as for the other usecase as a backend downloader for MlDatasets.jl etc)
If it is not working for this or if the code ends up ugly, then let me know.
(I am currently on holidays so am not responding too quick on Github right now, but I'll be back off next week)

@colbec
Copy link

colbec commented Mar 18, 2018

I'm getting a data error while trying to read the non-human images (source provided by @davidbp ) from the tutorial (see #48 (comment)). In trying to track down an alternate source of the data I came across a possibility at http://poggio-lab.mit.edu/PedestrianData.html however this yields a 404. I wrote to the contact mentioned on the lab page (dlees) but this address is no longer functioning.

@zygmuntszpak
Copy link
Member

You can download the original dataset by using the wayback machine. The following link appears to be downloading for me.

@colbec
Copy link

colbec commented Mar 18, 2018

@zygmuntszpak As far as I can see that link points to the human images, not the non_human. So the solution was to take the not_human images I have and resize them. The tutorial now runs to the end.

@davidbp
Copy link
Contributor Author

davidbp commented Mar 20, 2018

@zygmuntszpak The link works but it's not the same data. This link contains 924 images not 2500 images. Besides there are no Negative samples. I don't know where to download the exact same data from the tutorial.

@colbec colbec mentioned this issue Mar 20, 2018
Deepank308 added a commit to Deepank308/ImageFeatures.jl that referenced this issue Jan 6, 2019
The download link is now fixed for downloading pedestrian dataset as provided by @davidbp in JuliaImages#44.
The odd size of 65x129 is also fixed as reported by @colbec in JuliaImages#48.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

9 participants