PavlosMelissinos / enet-keras

A keras implementation of ENet (abandoned for the foreseeable future)
MIT License
115 stars 46 forks source link

Fix dependencies #33

Closed subhacom closed 4 years ago

subhacom commented 4 years ago

Pip throws and error on the line containing: gsutil=4.46 which seems to be a typo in requirements.txt . The same issue appears in environment.yml when using conda to install the dependencies. Other than that, many of these dependencies seem to be on out-dated versions. Is there any other recommended way to get the installation to work with more recent packages?

PavlosMelissinos commented 4 years ago

Yeah it's a typo, that should be a double equals sign in both files (i.e. gsutil==4.46), sorry about that.

It should be fine now but let me know if you encounter any problems.

Thanks for reporting the bug! :)

By the way, I'm really curious how you ended up here... ENet is a 2016 architecture and this repo has not really been active since 2017. I'd imagine there would be easier, better and faster ways to do semantic segmentation by now! What gives?

subhacom commented 4 years ago

Thanks for the prompt response, and for maintaining this project all these years.

I am really an outsider, just getting my feet wet with modern ML tools. Some sequence of google searches for image segmentation and keras landed me here. In particular, the enet paper shows up quite early in the search, and the authors' github page directs here for keras version.

PavlosMelissinos commented 4 years ago

Cool!

You might have heard this before but (in case you haven't) bear in mind that:

  1. Semantic image segmentation miiiight not be the best application of deep learning to get your feet wet with (in essence it's more granular/complex object detection, which is more granular image classification) and
  2. the project assumes that you have a decent GPU (for instance, a 8GB GTX1070 should be fine).

Let me know if there's anything else.

Cheers