I was reading this article (Faster R-CNN: Towards Real-Time Object Detection with Regional Proposal Network) and in yellow line:
What is the meaning of fine-tuned end-to-end?
Asked
Active
Viewed 922 times
3
-
Related: [Fine Tuning vs Joint Training vs Feature Extraction](http://stats.stackexchange.com/q/255364/12359) + [What does end to end mean in deep learning methods?](http://stats.stackexchange.com/q/224118/12359) – Franck Dernoncourt Feb 06 '17 at 15:29
-
@FranckDernoncourt It's not actual duplicate since I mean both end to end and fine tune terms – ᴀʀᴍᴀɴ Feb 06 '17 at 15:32
-
1the combination of both terms doesn't change the meaning of each term taken individually. – Franck Dernoncourt Feb 06 '17 at 19:37
-
@in duplicate question there is no define for fine-tuned – ᴀʀᴍᴀɴ Feb 06 '17 at 19:39
-
"You train a model on a dataset, use it for training on another dataset. This is fine tuning. [..] in the fine-tuning strategy all weights are changed when training on the new task" – Franck Dernoncourt Feb 06 '17 at 20:12
-
@I was looking for this answer , that user.. posted below , and it's not in duplicate question , that's why I said it's not duplicate – ᴀʀᴍᴀɴ Feb 06 '17 at 20:13
-
I see. Unfortunately Stack Exchange doesn't allow a user to mark a question as a duplicate of two other questions: [Question A is a duplicate of question B + question C: what to do?](http://meta.stats.stackexchange.com/q/4565/12359) – Franck Dernoncourt Feb 06 '17 at 20:15
1 Answers
1
My understanding of the sentence is: They started with a network (ImageNet) that had been trained on some other data set. They then trained this network on a new task ('region proposal') by feeding it examples from a new data set and adjusting the parameters to minimize the new loss function (i.e. using end-to-end training). Some more information about the term 'end-to-end training' can be found here and here.
