FAIMED 3D

use fastai to quickly train fully three-dimensional models on radiological data

Classification

from faimed3d.all import *

Load data in various medical formats (DICOM, NIfTI, NRRD) or even videos as simple as in fastai.

d = pd.read_csv('../data/radiopaedia_cases.csv')
dls = ImageDataLoaders3D.from_df(d,
                                 item_tfms = Resize3D((20, 112, 112)),
                                 batch_tfms = aug_transforms_3d(), 
                                 bs = 2, val_bs = 2)

Faimed3d provides multiple model architectures, pretrained on the UCF101 dataset for action recoginiton, which can be used for transfer learning.

Model	3-fold accuracy	duration/epoch	model size
efficientnet b0	92.5 %	9M:35S	48.8 MB
efficientnet b1	90.1 %	13M:20S	80.5 MB
resnet 18	87.6 %	6M:57S	339.1 MB
resnet 50	94.8 %	12M:16S	561.2 MB
resnet 101	96.0 %	17M:20S	1,030 MB

# slow
learn = cnn_learner_3d(dls, efficientnet_b0)

# slow
learn.lr_find()

SuggestedLRs(lr_min=0.014454397559165954, lr_steep=6.309573450380412e-07)

Click here for a more in-depth classification example.

Segmentation

dls = SegmentationDataLoaders3D.from_df(d,
                                 item_tfms = Resize3D((20, 112, 112)),
                                 batch_tfms = aug_transforms_3d(), 
                                 bs = 2, val_bs = 2)

All models in faimed3d can be used as a backbone for U-Nets, even with pre-trained weights.

# slow
learn = unet_learner_3d(dls, efficientnet_b0, n_out = 2)

# slow 
learn.lr_find()

SuggestedLRs(lr_min=0.33113112449646, lr_steep=0.10000000149011612)

Click here for a more in-depth segmentation example.

Extension to fastai for volumetric medical data

Related tags

Overview

FAIMED 3D

Classification

Segmentation

Owner

Keno

[CVPR 2021] "The Lottery Tickets Hypothesis for Supervised and Self-supervised Pre-training in Computer Vision Models" Tianlong Chen, Jonathan Frankle, Shiyu Chang, Sijia Liu, Yang Zhang, Michael Carbin, Zhangyang Wang

[ICCV 2021] Counterfactual Attention Learning for Fine-Grained Visual Categorization and Re-identification

Implementation of Cross Transformer for spatially-aware few-shot transfer, in Pytorch

Generating Fractals on Starknet with Cairo

ShapeGlot: Learning Language for Shape Differentiation

abess: Fast Best-Subset Selection in Python and R

Tensorflow implementation of Swin Transformer model.

PyTorch code for the ICCV'21 paper: "Always Be Dreaming: A New Approach for Class-Incremental Learning"

Code for our SIGCOMM'21 paper "Network Planning with Deep Reinforcement Learning".

Image inpainting using Gaussian Mixture Models

Implementation for paper "Towards the Generalization of Contrastive Self-Supervised Learning"

Prior-Guided Multi-View 3D Head Reconstruction

TDmatch is a Python library developed to perform matching tasks in three categories:

Meshed-Memory Transformer for Image Captioning. CVPR 2020

FastReID is a research platform that implements state-of-the-art re-identification algorithms.

A-ESRGAN aims to provide better super-resolution images by using multi-scale attention U-net discriminators.

Graph parsing approach to structured sentiment analysis.

Lunar is a neural network aimbot that uses real-time object detection accelerated with CUDA on Nvidia GPUs.

Designing a Minimal Retrieve-and-Read System for Open-Domain Question Answering (NAACL 2021)