Welcome to our Multi-Task Learning Archive Repository! This comprehensive collection houses a diverse range of cutting-edge multi-task learning algorithms, models, datasets, and research implementations. Whether you're a researcher, developer, or enthusiast exploring the realms of machine learning, this repository offers a rich resource pool to delve into multi-task learning techniques across various domains. Explore, experiment, and advance your understanding of simultaneous learning paradigms with our curated collection of resources and tools.
After cloning this repository, cd
inside and use the following commands to create a virtual environment
python -m venv .env
source .env/bin/activate
python -m pip install -U pip
The requirement packages for this repo are listed below, though pip install wheel
is recommended to run first
albumentations==1.3.1
fastparquet==2023.10.1
pandas==2.1.3
tensorboard==2.15.1
tqdm==4.66.1
wandb==0.16.0
scikit-learn==1.3.2
The final thing is to install Pytorch (This repository is tested using Ubuntu 22.04, CUDA 12.1, and NVIDIA driver 523s)
pip3 install torch torchvision torchaudio
Dataset | Mode | Status | Related Task |
---|---|---|---|
Oxford Pet III | - | Available | Segmentation (3 classes), Classification (37 classes) |
NYUV2 | - | Available | Segmentation (19 classes), Depth Estimation, Surface Normal |
Cityscape | fine | Available | Segmentation (19 classes), Depth Estimation |
Cityscape | coarse | Available | Segmentation (19 classes), Depth Estimation |
CelebA | - | Available | (40+) Attibute Classification (binary labelled), Deep Metric Learning (10k+ identity), Resconstruction (250k+ images), Disentanglement Learning |
Method | Code | Status |
---|---|---|
Gradient Normalization | gn | Available |
Uncertainty Weighting | uw | Available |
Dynamic Weight Average | dwa | Available |
Random Loss Weighting | rlw | Available |
MGDA | mgda | Available |
PCGRAD | pcgrad | Available |
CAGRAD | cagrad | Available |
Recon | recon | - |
NashMTL | nash | - |
Geometric Loss Strategy | geo | - |
Gradient Sign Dropout | gsd | - |
IMTL | imtl | - |
Gradient Vaccine | gvac | - |
MoCo | moco | - |
Aligned MTL | amtl | - |
Based Architecture | Mode | Dataset Available |
---|---|---|
Unet | Hard Parameter Sharing | OxfordPetIII |
SegNet | Hard Parameter Sharing | OxfordPetIII |
Using the parameter in main.py
to perform a customized training process. The experiment evaluation (i.e. loss value, metrics value) is recorded by toggling --log
and --wandb
.