sainathadapa / Kaggle Carvana Image Masking Challenge
Licence: mit
Top 15% ranked solution to the Carvana Image Masking Challenge on Kaggle
Stars: ✭ 13
Projects that are alternatives of or similar to Kaggle Carvana Image Masking Challenge
Pytorch Unet
Simple PyTorch implementations of U-Net/FullyConvNet (FCN) for image segmentation
Stars: ✭ 470 (+3515.38%)
Mutual labels: jupyter-notebook, semantic-segmentation
Deep Recommender System
深度学习在推荐系统中的应用及论文小结。
Stars: ✭ 657 (+4953.85%)
Mutual labels: kaggle, jupyter-notebook
Deeplearning
深度学习入门教程, 优秀文章, Deep Learning Tutorial
Stars: ✭ 6,783 (+52076.92%)
Mutual labels: kaggle, jupyter-notebook
Icnet Tensorflow
TensorFlow-based implementation of "ICNet for Real-Time Semantic Segmentation on High-Resolution Images".
Stars: ✭ 396 (+2946.15%)
Mutual labels: jupyter-notebook, semantic-segmentation
Face Mask Detection
Face Mask Detection system based on computer vision and deep learning using OpenCV and Tensorflow/Keras
Stars: ✭ 774 (+5853.85%)
Mutual labels: kaggle, jupyter-notebook
Steal
STEAL - Learning Semantic Boundaries from Noisy Annotations (CVPR 2019)
Stars: ✭ 424 (+3161.54%)
Mutual labels: jupyter-notebook, semantic-segmentation
Amazon Forest Computer Vision
Amazon Forest Computer Vision: Satellite Image tagging code using PyTorch / Keras with lots of PyTorch tricks
Stars: ✭ 346 (+2561.54%)
Mutual labels: kaggle, jupyter-notebook
Deeplab v3
Tensorflow Implementation of the Semantic Segmentation DeepLab_V3 CNN
Stars: ✭ 768 (+5807.69%)
Mutual labels: jupyter-notebook, semantic-segmentation
Pytorch 3dunet
3D U-Net model for volumetric semantic segmentation written in pytorch
Stars: ✭ 765 (+5784.62%)
Mutual labels: jupyter-notebook, semantic-segmentation
Vpgnet
VPGNet: Vanishing Point Guided Network for Lane and Road Marking Detection and Recognition (ICCV 2017)
Stars: ✭ 382 (+2838.46%)
Mutual labels: jupyter-notebook, semantic-segmentation
Deepfake Detection
DeepFake Detection: Detect the video is fake or not using InceptionResNetV2.
Stars: ✭ 23 (+76.92%)
Mutual labels: kaggle, jupyter-notebook
Pytorch Unet
PyTorch implementation of the U-Net for image semantic segmentation with high quality images
Stars: ✭ 4,770 (+36592.31%)
Mutual labels: kaggle, 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 (+3184.62%)
Mutual labels: jupyter-notebook, semantic-segmentation
Data Science Projects
DataScience projects for learning : Kaggle challenges, Object Recognition, Parsing, etc.
Stars: ✭ 361 (+2676.92%)
Mutual labels: kaggle, jupyter-notebook
Tutorials
CatBoost tutorials repository
Stars: ✭ 563 (+4230.77%)
Mutual labels: kaggle, jupyter-notebook
Cascaded Fcn
Source code for the MICCAI 2016 Paper "Automatic Liver and Lesion Segmentation in CT Using Cascaded Fully Convolutional NeuralNetworks and 3D Conditional Random Fields"
Stars: ✭ 296 (+2176.92%)
Mutual labels: jupyter-notebook, semantic-segmentation
Self Correction Human Parsing
An out-of-box human parsing representation extractor.
Stars: ✭ 319 (+2353.85%)
Mutual labels: jupyter-notebook, semantic-segmentation
This repository contains my personal code for the Kaggle competition - Carvana Image Masking Challenge.
| NBViewer Link | Description | Train set loss / dice coeff | Validation set loss / dice coeff | Kaggle Public LB Dice coeff |
|---|---|---|---|---|
| 000-general-viz-explore.ipynb | Exploration | - | - | - |
| 01-unet-128x128.ipynb | U-Net; 128x128; Contracting path filter sizes: 64, 128, 256, 512, 1024 | 0.0191; 0.9899 | 0.0130; 0.9939 | 0.9908 |
| 02-unet-256x256.ipynb | U-Net; 256x256; Contracting path filter sizes: 32, 64, 128, 256, 512, 1024 | 0.0122; 0.9936 | 0.0100; 0.9951 | 0.9942 |
| 03-unet-512x512.ipynb | U-Net; 512x512; Contracting path filter sizes: 16, 32, 64, 128, 256, 512, 1024 | 0.0095; 0.9950 | 0.0089; 0.9954 | 0.9958 |
| 04-unet-1024x1024.ipynb | U-Net; 1024x1024; Contracting path filter sizes: 8, 16, 32, 64, 128, 256, 512, 1024 | 0.0084; 0.9956 | 0.0085; 0.9956 | 0.9961 |
| 07-threshold-analysis.ipynb | Check to see if tuning the threshold value makes any difference to the score. | - | - | - |
| 08-bounding-boxes.ipynb | Using the model from Notebook 2, bounding boxes are first visualized, and then calculated for both train and test sets. | - | - | - |
| 09-full-res-model-vertical-cut-and-bbox.ipynb | U-Net; 1024x1024; Contracting path filter sizes: 8, 16, 32, 64, 128, 256, 512, 1024; Vertical height of the original image is first reduced to 1024 from 1280, using bounding boxes. The resulting 1918x1024 image is cut vertically (with some overlap) to two images, each of size 1024x1024. | 0.0071; 0.9967 | 0.0077; 0.9966 | 0.9967 |
| 10-model-predictions-viz.ipynb | Models from notebooks 3,4, 9 are compared. Also worst predictions from Notebook 9's model are visualized | - | - | - |
| 11-ensemble.ipynb | Ensemble of Notebook 3 and 9 models, by taking average of predictions from both models. | - | - | 0.9967 |
| 12-tiramisu-256.ipynb | One Hundred Layers Tiramisu; 256x256; Didn't fully train. Each epoch was taking way too long time to complete even for 256x256 resolution. | - | - | - |
| 13-bounding-boxes-revisit.ipynb | Boundary boxes are analysed further | - | - | - |
| 14-full-res-model-1920-1080.ipynb | U-Net; 1920x1280; Contracting path filter sizes: 8, 16, 32, 64, 128, 256, 512, 1024; The output from Notebook 3's model is added as an additional depth layer, after resizing to full resolution. | 0.0030; 0.9986 | 0.0139; 0.9958 | - |
| 16-unet-inception.ipynb | UNet's VGG style Conv-BN-Act layers are replaced by InceptionV3 style blocks | 0.0184; 0.9905 | 0.0163; 0.9915 | - |
| 17-full-res-model-angle-09-bbox.ipynb | U-Net; 1280x896; Contracting path filter sizes: 8, 16, 32, 64, 128, 256, 512, 1024; Original images are cut to 1280x896 from 1918x1280, using bounding boxes. | 0.0072; 0.9969 | 0.0120; 0.9957 | - |
| 18-full-res-model-angle-09-bbox-no-maxpool.ipynb | Replaced max pooling from the above model, with 2x2 Convolution layer with 2x2 stride | 0.0283; 0.9880 | 0.0417; 0.9844 | - |
| 19-full-res-model-all-angles-vertical-cut-no-bbox.ipynb | The original 1918x1280 image is cut into two halves vertically (with some overlap), each with size 1024x1280. | 0.0064; 0.9968 | 0.0069; 0.9966 | 0.9967 |
| 20-full-res-model-all-angles-horizontal-cut-no-bbox.ipynb | The original 1918x1280 image is cut into two halves horizontally (with some overlap), each with size 1920x768. | - | - | 0.9967 |
| 21-ensemble.ipynb | Ensemble of models from Notebooks 19 and 20 (averaged the predictions) | - | - | 0.9968 |
| 22-unet-different-initial-filters.ipynb | U-net architecture with more filters in the first layer | - | - | - |
The Private Leaderboard score for the Ensemble model was 0.9965.
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].