Remove opencv dependence about caffe HOT 3 CLOSED

The detection fine-tuning code, which I'm planning to merge into core caffe
in the near future, uses some more features from opencv (crop a ROI and
image resizing). Even though it's a heavy dependency, I can imagine that
having it around (or some other image manipulation library) is going to be
very useful to doing more complex transformations of data on-the-fly while
training.

On Tue, Jan 21, 2014 at 1:52 PM, Evan Shelhamer [email protected]:

opencv is a heavy dependency when we are only using it to do image IO. The
upfront install work could be lessened if there were simple libpng and
libjpeg wrappers to handle the image loading and saving.

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/45
.

http://www.cs.berkeley.edu/~rbg/

from caffe.

Second Ross's opinion - I am not particularly worried about opencv at this
stage, since it seems reasonable for one to ask vision researchers to have
opencv installed... Having a duplicate wrapper around e.g. libjpeg instead
of using the opencv interface seems a little bit unnecessary, since we are
already heavily using other libraries like leveldb, protobuf, and most
notably boost :)

Yangqing

On Tue, Jan 21, 2014 at 1:56 PM, Ross Girshick [email protected]:

The detection fine-tuning code, which I'm planning to merge into core
caffe
in the near future, uses some more features from opencv (crop a ROI and
image resizing). Even though it's a heavy dependency, I can imagine that
having it around (or some other image manipulation library) is going to be
very useful to doing more complex transformations of data on-the-fly while
training.

On Tue, Jan 21, 2014 at 1:52 PM, Evan Shelhamer [email protected]:

opencv is a heavy dependency when we are only using it to do image IO.
The
upfront install work could be lessened if there were simple libpng and
libjpeg wrappers to handle the image loading and saving.

—
Reply to this email directly or view it on GitHub<
https://github.com/BVLC/caffe/issues/45>
.

http://www.cs.berkeley.edu/~rbg/

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/45#issuecomment-32968210
.

from caffe.

True, everyone in vision most likely has opencv anyway and boost is no small dependency.

from caffe.