Official Pytorch implementation of Meta Internal Learning

Last update: Aug 24, 2022

Related tags

Overview

Meta Internal Learning

This repository is the official implementation of Meta Internal Learning by Raphael Bensadoun, Shir Gur, Tomer Galanti, Lior Wolf.

Project | arXiv | Code

Requirements

To install requirements:

pip install -r requirements.txt

Training on a dataset

Create a folder X containing the images. (see structure in data folder)
Determine how many iterations Y to train by scale (depends mostly on the size of the dataset, you may refer to the appendix for reference).
Run

python train.py --image-path X --batch-size 16 --visualize --niter Y --min-size 25 --checkname X_result --SAVE-MODEL --SAVE-IMGS

Generated images, tensorboard logs and trained models are stored in MetaInternalLearning/run/X_result.

The default input format is jpg. Use '--file-suffix png' for .png files.

Examples from paper -

Places-50 -

python train.py --image-path data/places_50 --batch-size 16 --visualize --niter 4000  --min-size 28 --checkname places_50_result --SAVE-MODEL --SAVE-IMGS

LSUN-50

python train.py --image-path data/lsun_50 --batch-size 16 --visualize --niter 5000  --checkname lsun_50_result --SAVE-MODEL --SAVE-IMGS

Valley dataset can be downloaded here - http://places2.csail.mit.edu/download.html (256x256 small images) and can be divided into subsets as mentioned in the paper.

V500 -

python train.py --image-path data/V500 --batch-size 16 --visualize --niter 25000 --min-size 25 --checkname v500_result --ar 1 --SAVE-MODEL

V2500 -

python train_dataset_parallel.py --image-path data/V2500 --batch-size 16 --niter 100000 --rec-weight 50 --min-size 25 --checkname v2500_result --ar 1 --SAVE-MODEL

V5000 -

python train_dataset_parallel.py --image-path data/V5000 --batch-size 16 --niter 150000 --rec-weight 50 --min-size 25 --checkname v5000_result --ar 1 --SAVE-MODEL

Applications

Coming soon!

Official Pytorch implementation of Meta Internal Learning

Related tags

Overview

Meta Internal Learning

Requirements

Training on a dataset

Examples from paper -

Applications

Owner

A proof of concept ai-powered Recaptcha v2 solver

Everything you want about DP-Based Federated Learning, including Papers and Code. (Mechanism: Laplace or Gaussian, Dataset: femnist, shakespeare, mnist, cifar-10 and fashion-mnist. )

Language Models for the legal domain in Spanish done @ BSC-TEMU within the "Plan de las Tecnologías del Lenguaje" (Plan-TL).

A very simple tool to rewrite parameters such as attributes and constants for OPs in ONNX models. Simple Attribute and Constant Modifier for ONNX.

MolRep: A Deep Representation Learning Library for Molecular Property Prediction

A machine learning library for spiking neural networks. Supports training with both torch and jax pipelines, and deployment to neuromorphic hardware.

Решения, подсказки, тесты и утилиты для тренировки по алгоритмам от Яндекса.

Code and models for ICCV2021 paper "Robust Object Detection via Instance-Level Temporal Cycle Confusion".

Code for our ACL 2021 paper "One2Set: Generating Diverse Keyphrases as a Set"

Official PyTorch implementation of StyleGAN3

Perturbed Self-Distillation: Weakly Supervised Large-Scale Point Cloud Semantic Segmentation (ICCV2021)

Official implementation of Representer Point Selection via Local Jacobian Expansion for Post-hoc Classifier Explanation of Deep Neural Networks and Ensemble Models at NeurIPS 2021

PyTorch implementation of the paper: Label Noise Transition Matrix Estimation for Tasks with Lower-Quality Features

Lane follower: Lane-detector (OpenCV) + Object-detector (YOLO5) + CAN-bus

meProp: Sparsified Back Propagation for Accelerated Deep Learning (ICML 2017)

Deep Learning Tutorial for Kaggle Ultrasound Nerve Segmentation competition, using Keras

PointNetVLAD: Deep Point Cloud Based Retrieval for Large-Scale Place Recognition, CVPR 2018

Hooks for VCOCO

Implementation of Neural Distance Embeddings for Biological Sequences (NeuroSEED) in PyTorch

Learning Tracking Representations via Dual-Branch Fully Transformer Networks