Excuse me does anybody know how to transfer learning a Faster RCNN to my own dataset? Or is there any reference for that? because I still don’t understand is it included as my base network in my model or what. Thank you very much
2 Likes
How about the following, though I’ve not tested this yet:
-
In TensorFlow Hub (a repository of pre-trained TensorFlow models), click on/search for “object detection” models: (link to Image Object Detection results: Find Pre-trained Models | Kaggle)
-
In the results, there is
faster_rcnn/openimages_v4/inception_resnet_v2(link: Google | faster_rcnn/inception_resnet_v2 | Kaggle)Object detection model trained on Open Images V4 with ImageNet pre-trained Inception Resnet V2 as image feature extractor.
FasterRCNN+InceptionResNetV2 network trained on Open Images V4.
-
-
Then, check out the Transfer learning with TensorFlow Hub | TensorFlow Core tutorial.
Looping in @lgusm (DevRel) and the docs team - notebook co-authors @billy @markdaoust
1 Like
TensorFlow HUB currently doesn’t support fine-tuning on Faster RCNN and generally on other models in the Object Detection API.
But it is a quite requested feature. See Make Object detection models fine-tunable · Issue #678 · tensorflow/hub · GitHub.
Let’s see if there is any other progress to share on this.
2 Likes
Might be available (no deadline I guess) under TensorFlow Model Garden:
2 Likes