Note CPU performance insufficent?

[Matthias84]

Describe the feature you'd like
Add techniques and docs that help people without adequate Nvidia GPU to get started with detecto to fit custom models.
Currently you see only @epoch 1...@ but don't see any progress even in hours or running all night long.

Describe the use cases of the feature
Detecto is considered as high level abstraction to simplify getting working models and to train models.
Thus users are usually not familar with complex installation process of CUDA / don't own a supported NVIDIA card / ...
But if you currently start training a custom model, there is nothing more than a hint, that it will take long time on CPU only. This results in frustration, as the simplified process now gets suddenly more complex. This isn't a easy start to the world of ML for object detection.

Additional context
I suggest to give the users more assistance on how to start custom model training, even if they use only a CPU

• Show more logging during training
• check if training material is performant (resolution, size , size validation set)
• estimate rough training duration based on number of images
• document experiences (me: Training on i7 6600 16GB RAM, win10 has never finished epoch 1 for >8hrs
• tipps on general speedup (shrinked size, resolution, ...)

[alankbi]

Thanks for the suggestion - I incorporated parts of this in the release of v1.2.0. In particular, it now shows you the progress through each epoch and gives users a warning if they're using a CPU.

Unfortunately, I don't think Detecto (or even just object detection models in general) will ever run in a reasonable amount of time on a normal CPU in the foreseeable future, so I'd definitely recommend checking out Colab or something similar for a free GPU if you don't have one.