Can anyone suggest a good example for one shot learning model ?
1 Like
Bhack
July 13, 2021, 4:05pm
#3
This will give you a good overview of the topic in different domains:
https://arxiv.org/abs/1904.05046
Then probably you could find something more specific for your application target.
Given that Siamese networks can work for one-shot learning (such as image recognition (Koch, Zemel & Salakhutdinov (2015) , Jadon (2020) ), maybe you could try adapting these Siamese net examples from the Keras community @fchollet :
Perhaps, @Sayak_Paul knows more about this.
In case you’re interested in few-shot learning examples, which are somewhat related:
MAML (model-agnostic meta-learning): tensorflow-maml/maml.ipynb at master · hereismari/tensorflow-maml · GitHub
(from the MAML paper by Finn, Abbeel & Levine (2017) )
The Keras community created an implementation of a MAML-inspired method called Reptile here: Few-Shot learning with Reptile (cc @fchollet )
(from the Reptile paper by Nichol, Achiam & Schulman (2018) ).
Bhack
July 13, 2021, 7:33pm
#6
I don’t know your domain but e.g. if you want an high level API example for the object detection API in model garden we have:
{
"cells": [
{
"cell_type": "markdown",
"metadata": {
"colab_type": "text",
"id": "rOvvWAVTkMR7"
},
"source": [
"# Eager Few Shot Object Detection Colab\n",
"\n",
"Welcome to the Eager Few Shot Object Detection Colab --- in this colab we demonstrate fine tuning of a (TF2 friendly) RetinaNet architecture on very few examples of a novel class after initializing from a pre-trained COCO checkpoint.\n",
"Training runs in eager mode.\n",
"\n",
"Estimated time to run through this colab (with GPU): \u003c 5 minutes."
]
},
{
"cell_type": "markdown",
"metadata": {
show original
For TFLite
{
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "rOvvWAVTkMR7"
},
"source": [
"# Introduction\n",
"\n",
"Welcome to the **Few Shot Object Detection for TensorFlow Lite** Colab. Here, we demonstrate fine tuning of a SSD architecture (pre-trained on COCO) on very few examples of a *novel* class. We will then generate a (downloadable) TensorFlow Lite model for on-device inference.\n",
"\n",
"**NOTE:** This Colab is meant for the few-shot detection use-case. To train a model on a large dataset, please follow the [TF2 training](https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/tf2_training_and_evaluation.md#training) documentation and then [convert](https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/running_on_mobile_tf2.md) the model to TensorFlow Lite."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "3U2sv0upw04O"
},
show original
1 Like
