bfortuner / Pytorch_tiramisu
Licence: mit
FC-DenseNet in PyTorch for Semantic Segmentation
Stars: ✭ 267
Projects that are alternatives of or similar to Pytorch tiramisu
Deeplabv3 Plus
Tensorflow 2.3.0 implementation of DeepLabV3-Plus
Stars: ✭ 32 (-88.01%)
Mutual labels: jupyter-notebook, semantic-segmentation
Multiclass Semantic Segmentation Camvid
Tensorflow 2 implementation of complete pipeline for multiclass image semantic segmentation using UNet, SegNet and FCN32 architectures on Cambridge-driving Labeled Video Database (CamVid) dataset.
Stars: ✭ 67 (-74.91%)
Mutual labels: jupyter-notebook, semantic-segmentation
Pytorch 3dunet
3D U-Net model for volumetric semantic segmentation written in pytorch
Stars: ✭ 765 (+186.52%)
Mutual labels: jupyter-notebook, semantic-segmentation
Fcn For Semantic Segmentation
Implemention of FCN-8 and FCN-16 with Keras and uses CRF as post processing
Stars: ✭ 155 (-41.95%)
Mutual labels: jupyter-notebook, semantic-segmentation
Deeplab v3
Tensorflow Implementation of the Semantic Segmentation DeepLab_V3 CNN
Stars: ✭ 768 (+187.64%)
Mutual labels: jupyter-notebook, semantic-segmentation
Spacenet building detection
Project to train/test convolutional neural networks to extract buildings from SpaceNet satellite imageries.
Stars: ✭ 83 (-68.91%)
Mutual labels: jupyter-notebook, semantic-segmentation
Icnet Tensorflow
TensorFlow-based implementation of "ICNet for Real-Time Semantic Segmentation on High-Resolution Images".
Stars: ✭ 396 (+48.31%)
Mutual labels: jupyter-notebook, semantic-segmentation
Multi Task Refinenet
Multi-Task (Joint Segmentation / Depth / Surface Normas) Real-Time Light-Weight RefineNet
Stars: ✭ 139 (-47.94%)
Mutual labels: jupyter-notebook, semantic-segmentation
Unet
Generic U-Net Tensorflow 2 implementation for semantic segmentation
Stars: ✭ 100 (-62.55%)
Mutual labels: jupyter-notebook, semantic-segmentation
Pytorch Unet
Simple PyTorch implementations of U-Net/FullyConvNet (FCN) for image segmentation
Stars: ✭ 470 (+76.03%)
Mutual labels: jupyter-notebook, semantic-segmentation
Kili Playground
Simplest and fastest image and text annotation tool.
Stars: ✭ 166 (-37.83%)
Mutual labels: jupyter-notebook, semantic-segmentation
Probabilistic unet
A U-Net combined with a variational auto-encoder that is able to learn conditional distributions over semantic segmentations.
Stars: ✭ 427 (+59.93%)
Mutual labels: jupyter-notebook, semantic-segmentation
Kaggle Carvana Image Masking Challenge
Top 15% ranked solution to the Carvana Image Masking Challenge on Kaggle
Stars: ✭ 13 (-95.13%)
Mutual labels: jupyter-notebook, semantic-segmentation
Steal
STEAL - Learning Semantic Boundaries from Noisy Annotations (CVPR 2019)
Stars: ✭ 424 (+58.8%)
Mutual labels: jupyter-notebook, semantic-segmentation
Unet Tgs
Applying UNET Model on TGS Salt Identification Challenge hosted on Kaggle
Stars: ✭ 81 (-69.66%)
Mutual labels: jupyter-notebook, semantic-segmentation
Self Correction Human Parsing
An out-of-box human parsing representation extractor.
Stars: ✭ 319 (+19.48%)
Mutual labels: jupyter-notebook, semantic-segmentation
Vpgnet
VPGNet: Vanishing Point Guided Network for Lane and Road Marking Detection and Recognition (ICCV 2017)
Stars: ✭ 382 (+43.07%)
Mutual labels: jupyter-notebook, semantic-segmentation
Deep Residual Unet
ResUNet, a semantic segmentation model inspired by the deep residual learning and UNet. An architecture that take advantages from both(Residual and UNet) models.
Stars: ✭ 97 (-63.67%)
Mutual labels: jupyter-notebook, semantic-segmentation
Keras Segmentation Deeplab V3.1
An awesome semantic segmentation model that runs in real time
Stars: ✭ 156 (-41.57%)
Mutual labels: jupyter-notebook, semantic-segmentation
One Hundred Layers Tiramisu
PyTorch implementation of The One Hundred Layers Tiramisu: Fully Convolutional DenseNets for Semantic Segmentation.
Tiramisu combines DensetNet and U-Net for high performance semantic segmentation. In this repository, we attempt to replicate the authors' results on the CamVid dataset.
Setup
Requires Anaconda for Python3 installed.
conda create --name tiramisu python=3.6
source activate tiramisu
conda install pytorch torchvision -c pytorch
The train.ipynb notebook shows a basic train/test workflow.
Dataset
Download
Specs
- Training: 367 frames
- Validation: 101 frames
- TestSet: 233 frames
- Dimensions: 360x480
- Classes: 11 (+1 background)
Architecture
Tiramisu adopts the UNet design with downsampling, bottleneck, and upsampling paths and skip connections. It replaces convolution and max pooling layers with Dense blocks from the DenseNet architecture. Dense blocks contain residual connections like in ResNet except they concatenate, rather than sum, prior feature maps.
Layers
FCDenseNet103
Authors' Results
Our Results
FCDenseNet67
We trained for 670 epochs (224x224 crops) with 100 epochs fine-tuning (full-size images). The authors mention "global accuracy" of 90.8 for FC-DenseNet67 on Camvid, compared to our 86.8. If we exclude the 'background' class, accuracy increases to ~89%. We think the authors did this, but haven't confirmed.
| Dataset | Loss | Accuracy |
|---|---|---|
| Validation | .209 | 92.5 |
| Testset | .435 | 86.8 |
FCDenseNet103
We trained for 874 epochs with 50 epochs fine-tuning.
| Dataset | Loss | Accuracy |
|---|---|---|
| Validation | .178 | 92.8 |
| Testset | .441 | 86.6 |
Predictions
Training
Hyperparameters
- WeightInitialization = HeUniform
- Optimizer = RMSProp
- LR = .001 with exponential decay of 0.995 after each epoch
- Data Augmentation = Random Crops, Vertical Flips
- ValidationSet with early stopping based on IoU or MeanAccuracy with patience of 100 (50 during finetuning)
- WeightDecay = .0001
- Finetune with full-size images, LR = .0001
- Dropout = 0.2
- BatchNorm "we use current batch stats at training, validation, and test time"
References and Links
Note that the project description data, including the texts, logos, images, and/or trademarks,
for each open source project belongs to its rightful owner.
If you wish to add or remove any projects, please contact us at [email protected].