3d Cnn Action Recognition
| Project Name | Stars | Downloads | Repos Using This | Packages Using This | Most Recent Commit | Total Releases | Latest Release | Open Issues | License | Language |
|---|---|---|---|---|---|---|---|---|---|---|
| Mmdetection3d | 3,402 | a day ago | 25 | June 14, 2022 | 240 | apache-2.0 | Python | |||
| OpenMMLab's next-generation platform for general 3D object detection. | ||||||||||
| Openpcdet | 3,399 | 5 days ago | 48 | apache-2.0 | Python | |||||
| OpenPCDet Toolbox for LiDAR-based 3D Object Detection. | ||||||||||
| Nnunet | 3,390 |  | 6 | a day ago | 1 | May 28, 2021 | 250 | apache-2.0 | Python | |
| Objectron | 1,958 | 8 months ago | 21 | other | Jupyter Notebook | |||||
| Objectron is a dataset of short, object-centric video clips. In addition, the videos also contain AR session metadata including camera poses, sparse point-clouds and planes. In each video, the camera moves around and above the object and captures it from different views. Each object is annotated with a 3D bounding box. The 3D bounding box describes the object’s position, orientation, and dimensions. The dataset contains about 15K annotated video clips and 4M annotated images in the following categories: bikes, books, bottles, cameras, cereal boxes, chairs, cups, laptops, and shoes | ||||||||||
| 3d Pointcloud | 1,374 | 4 days ago | 2 | Python | ||||||
| Papers and Datasets about Point Cloud. | ||||||||||
| Tutorials | 1,030 | a day ago | 54 | apache-2.0 | Jupyter Notebook | |||||
| MONAI Tutorials | ||||||||||
| Medmnist | 701 | 2 months ago | 3 | May 06, 2022 | 2 | apache-2.0 | Python | |||
| 18 MNIST-like Datasets for 2D and 3D Biomedical Image Classification: pip install medmnist | ||||||||||
| 2d And 3d Face Alignment | 693 | 5 years ago | 4 | bsd-3-clause | Lua | |||||
| This repository implements a demo of the networks described in "How far are we from solving the 2D & 3D Face Alignment problem? (and a dataset of 230,000 3D facial landmarks)" paper. | ||||||||||
| Curated List Of Awesome 3d Morphable Model Software And Data | 664 | 8 months ago | ||||||||
| The idea of this list is to collect shared data and algorithms around 3D Morphable Models. You are invited to contribute to this list by adding a pull request. The original list arised from the Dagstuhl seminar on 3D Morphable Models https://www.dagstuhl.de/19102 in March 2019. | ||||||||||
| 3dmppe_posenet_release | 616 | 9 months ago | 23 | mit | Python | |||||
| Official PyTorch implementation of "Camera Distance-aware Top-down Approach for 3D Multi-person Pose Estimation from a Single RGB Image", ICCV 2019 | ||||||||||
3D - Convolutional Neural Network For Action Recognition
Inplementation of 3D Convolutional Neural Network for video classification using Keras(with tensorflow as backend).
Description
This code requires UCF-101 dataset. This code generates graphs of accuracy and loss, plot of model, result and class names as txt file and model as hd5 and json.
You can use visualize_input.py to make an input image which will maximize the specific output. This code is able to maximize a layer's output of any classification model. (Only dense layer convolutional layer(2D/3D) and pooling layer(2D/3D) are allowed.)
Requirements
python3
opencv3 (with ffmpeg), keras, numpy, tqdm
Options
Options of 3dcnn.py are as following:
--batch batch size, default is 128
--epoch the number of epochs, default is 100
--videos a name of directory where dataset is stored, default is UCF101
--nclass the number of classes you want to use, default is 101
--output a directory where the results described above will be saved
--color use RGB image or grayscale image, default is False
--skip get frames at interval or contenuously, default is True
--depth the number of frames to use, default is 10
Options of 3dcnn_ensemble.py are almost same as those of 3dcnn.py.
You can use --nmodel option to set the number of models.
Options of visualize_input.py are as follows:
--model saved json file of a model
--weights saved hd5 file of a model weights
--layernames True to show layer names of a model, default is False
--name the name of a layer which will be maximized
--index the index of a layer output which will be maximized
--iter the number of iteration, default is 20
You can see more information by using --help option
Demo
You can execute like the following:
python 3dcnn.py --batch 32 --epoch 50 --videos dataset/ --nclass 10 --output 3dcnnresult/ --color True --skip False --depth 10
You can generate the input image which maximizes 0th output of layer named 'dense_2' like this:
python visualize_input.py -m result_cnn_10class/ucf101cnnmodel.json -w result_cnn_10class/ucf101cnnmodel.hd5 -n 'dense_2' -i 0 --iter 100
Other files
3dcnn-my-model.py 3DCNN newly designed model (check structure). Accuracy - 61%
2dcnn.py 2DCNN model
display.py get example images from the dataset.
videoto3d.py get frames from a video, extract a class name from filename of a video in UCF101.
