Hi Olmo,
On 30. Oct 2018, at 06:32, Olmo Chiara <olmo.chiara@gmail.com> wrote:
3. Any update with respect to the pedestrian dataset?
I did talk with Mr. Garcke and Bastian about this. As you might know, the problem with the current pedestrian dataset is, that the specific subset we used has license issues. The original Daimler dataset is freely available, our subset isn’t. For legal reasons we need a freely available dataset. I would like to ask you to create a new subset/dataset which we can use for the sheet. You can either use the original Daimler datasets and create a new subset with identical parameters (image size, number of images etc), or look into one of the newer pedestrian classification datasets, which might be more interesting. The survey referenced on the sheet mentions a few newer datasets, I think all of them are in color. It is also easy to find pedestrian datasets on Google. You would have to pick a reasonable number of images of reasonable size, run the tasks of the sheet on them and check if everything stills works in reasonable time. You can decide if you want to adapt the tasks for colored images or greyscale them beforehand, and how large the dataset should be. Regarding the HoG features, I think the C++ implementation can already handle colored input, but the binding might need some modification in order to use this. Kind regards, Jannik