41 skills found · Page 1 of 2
raoyongming / CAL[ICCV 2021] Counterfactual Attention Learning for Fine-Grained Visual Categorization and Re-identification
mooch443 / TrexTRex, a fast multi-animal tracking system with markerless identification, and 2D estimation of posture and visual fields.
wh200720041 / SRLCDfast loop closure detection (online visual place recognition) via saliency re-identification IROS 2020
ustctug / Ustclogo中国科学技术大学视觉识别素材包. USTC Visual Identification Pack for LaTeX users, with ease of using themed colors
CQUtug / CQULogo重庆大学视觉标识素材包 --- Visual Identification Bundle of Chongqing University
CWOA / MetricLearningIdentificationUnderpinning code for our paper - "Visual Identification of Individual Holstein Friesian Cattle via Deep Metric Learning"
mukeshmithrakumar / RetinaNetKeras implementation of RetinaNet for object detection and visual relationship identification
yizhuoyang / AV FDTIAV-FDTI: Audio-Visual Fusion for Drone Threat Identification
ajaybhatiya1234 / DEEP FACE Dectection01 Read the technical deep dive: https://www.dessa.com/post/deepfake-detection-that-actually-works # Visual DeepFake Detection In our recent [article](https://www.dessa.com/post/deepfake-detection-that-actually-works), we make the following contributions: * We show that the model proposed in current state of the art in video manipulation (FaceForensics++) does not generalize to real-life videos randomly collected from Youtube. * We show the need for the detector to be constantly updated with real-world data, and propose an initial solution in hopes of solving deepfake video detection. Our Pytorch implementation, conducts extensive experiments to demonstrate that the datasets produced by Google and detailed in the FaceForensics++ paper are not sufficient for making neural networks generalize to detect real-life face manipulation techniques. It also provides a current solution for such behavior which relies on adding more data. Our Pytorch model is based on a pre-trained ResNet18 on Imagenet, that we finetune to solve the deepfake detection problem. We also conduct large scale experiments using Dessa's open source scheduler + experiment manger [Atlas](https://github.com/dessa-research/atlas). ## Setup ## Prerequisities To run the code, your system should meet the following requirements: RAM >= 32GB , GPUs >=1 ## Steps 0. Install [nvidia-docker](https://github.com/nvidia/nvidia-docker/wiki/Installation-(version-2.0)) 00. Install [ffmpeg](https://www.ffmpeg.org/download.html) or `sudo apt install ffmpeg` 1. Git Clone this repository. 2. If you haven't already, install [Atlas](https://github.com/dessa-research/atlas). 3. Once you've installed Atlas, activate your environment if you haven't already, and navigate to your project folder. That's it, You're ready to go! ## Datasets Half of the dataset used in this project is from the [FaceForensics](https://github.com/ondyari/FaceForensics/tree/master/dataset) deepfake detection dataset. . To download this data, please make sure to fill out the [google form](https://github.com/ondyari/FaceForensics/#access) to request access to the data. For the dataset that we collected from Youtube, it is accessible on [S3](ttps://deepfake-detection.s3.amazonaws.com/augment_deepfake.tar.gz) for download. To automatically download and restructure both datasets, please execute: ``` bash restructure_data.sh faceforensics_download.py ``` Note: You need to have received the download script from FaceForensics++ people before executing the restructure script. Note2: We created the `restructure_data.sh` to do a split that replicates our exact experiments avaiable in the UI above, please feel free to change the splits as you wish. ## Walkthrough Before starting to train/evaluate models, we should first create the docker image that we will be running our experiments with. To do so, we already prepared a dockerfile to do that inside `custom_docker_image`. To create the docker image, execute the following commands in terminal: ``` cd custom_docker_image nvidia-docker build . -t atlas_ff ``` Note: if you change the image name, please make sure you also modify line 16 of `job.config.yaml` to match the docker image name. Inside `job.config.yaml`, please modify the data path on host from `/media/biggie2/FaceForensics/datasets/` to the absolute path of your `datasets` folder. The folder containing your datasets should have the following structure: ``` datasets ├── augment_deepfake (2) │ ├── fake │ │ └── frames │ ├── real │ │ └── frames │ └── val │ ├── fake │ └── real ├── base_deepfake (1) │ ├── fake │ │ └── frames │ ├── real │ │ └── frames │ └── val │ ├── fake │ └── real ├── both_deepfake (3) │ ├── fake │ │ └── frames │ ├── real │ │ └── frames │ └── val │ ├── fake │ └── real ├── precomputed (4) └── T_deepfake (0) ├── manipulated_sequences │ ├── DeepFakeDetection │ ├── Deepfakes │ ├── Face2Face │ ├── FaceSwap │ └── NeuralTextures └── original_sequences ├── actors └── youtube ``` Notes: * (0) is the dataset downloaded using the FaceForensics repo scripts * (1) is a reshaped version of FaceForensics data to match the expected structure by the codebase. subfolders called `frames` contain frames collected using `ffmpeg` * (2) is the augmented dataset, collected from youtube, available on s3. * (3) is the combination of both base and augmented datasets. * (4) precomputed will be automatically created during training. It holds cashed cropped frames. Then, to run all the experiments we will show in the article to come, you can launch the script `hparams_search.py` using: ```bash python hparams_search.py ``` ## Results In the following pictures, the title for each subplot is in the form `real_prob, fake_prob | prediction | label`. #### Model trained on FaceForensics++ dataset For models trained on the paper dataset alone, we notice that the model only learns to detect the manipulation techniques mentioned in the paper and misses all the manipulations in real world data (from data)   #### Model trained on Youtube dataset Models trained on the youtube data alone learn to detect real world deepfakes, but also learn to detect easy deepfakes in the paper dataset as well. These models however fail to detect any other type of manipulation (such as NeuralTextures).   #### Model trained on Paper + Youtube dataset Finally, models trained on the combination of both datasets together, learns to detect both real world manipulation techniques as well as the other methods mentioned in FaceForensics++ paper.   for a more in depth explanation of these results, please refer to the [article](https://www.dessa.com/post/deepfake-detection-that-actually-works) we published. More results can be seen in the [interactive UI](http://deepfake-detection.dessa.com/projects) ## Help improve this technology Please feel free to fork this work and keep pushing on it. If you also want to help improving the deepfake detection datasets, please share your real/forged samples at foundations@dessa.com. ## LICENSE © 2020 Square, Inc. ATLAS, DESSA, the Dessa Logo, and others are trademarks of Square, Inc. All third party names and trademarks are properties of their respective owners and are used for identification purposes only.
Satya3720 / Rock Identification Using Deep Convolution Neural NetworkRocks are a fundamental component of Earth. The automatic identification of rock type in the field would aid geological surveying, education, and automatic mapping. It is a basic part of geological surveying and research, and mineral resources exploration. The automatic identification of rock type in the field would aid geological surveying, education, and automatic mapping. Working conditions in the field generally limit identification to visual methods, including using a magnifying glass for fine-grained rocks. Visual inspection assesses properties such as colour, composition, grain size, and structure. The attributes of rocks reflect their mineral and chemical composition, formation environment, and genesis. The colour of rock reflects its chemical composition. But these analysis is time taken process to identify the rocks.Its application here has effectively identified rock types from images captured in the field. This paper proposes an accurate approach for identifying rock types in the field based on image analysis using deep convolutional neural networks. Solution: Deep learning is receiving significant research attention for pattern recognition and machine learning. Its application here has effectively identified rock types from images captured in the field. This paper proposes an accurate approach for identifying rock types in the field based on image analysis using deep convolutional neural networks. The results show that the proposed approach based on deep learning represents an improvement in intelligent rock-type identification and solves several difficulties facing the automated identification of rock types in the field.Who are experienced in the field of geological they can identify the rocks easily. But who are new to the field, it can help to identify the type of rock.
rika1024 / JVTCJoint Visual and Temporal Consistency for Unsupervised Domain Adaptive Person Re-Identification ECCV 2020
moravianlibrary / DifferDeveloping preservation processes for a trusted digital repository requires the utilization of new methods and technologies, which have helped to accelerate the whole process of control. The current approach at the Digital Preservation Standards Department at The National Library of the Czech Republic is to develop a quality control application for still image file formats capable of performing identification, characterization, validation and visual/mathematical comparison integrated into an operational digital preservation framework. The online application DIFFER is utilizing existing tools (JHOVE, FITS, ExifTool, KDU_expand, DJVUDUMP, Jpylyzer, etc.), which are mainly used separately across a whole spectrum of existing projects. This open source application comes with a well-structured and uniform GUI, which helps the user to understand the relationships between various file format properties, detect visual and non-visual errors and simplifies decision-making. An additional feature called compliance-check is designed to help us check the required specifications of the JPEG2000 file format.
linh-gist / VisualRFSThis is the official Python and C++ implementation repository for a paper entitled "Visual multi-object tracking with re-identification and occlusion handling using labeled random finite sets", Pattern Recognition (https://arxiv.org/pdf/2407.08872).
GayatriBehara / Face And Facial Expression Detection Based Authenticated ZigBee Based RoboIn this project a robot that can be operated by authorized person or operator is implemented. For this purpose we use a face recognition system which is capable of identifying the authorized person which allows him to command and operate it. The face recognition system consists of a web based camera which captures the image of human and this image is processed in MATLAB software. After processing the image it generates the activation code for the robot to be operated. The hardware system is based on the ATMEL microcontroller and an Zigbee module. The system provides continuous visual monitoring through the small camera attached to the mobile robot, sending data to the control unit when necessary. Remote testing is done on the mobile robot for search and rescue missions via an established radio frequency (RF) communication using DIGI XBee RF module. Intelligent mobile robots and cooperative multi agent robotic systems can be very efficient tools to speed up search and research operations in remote areas. This prototype robot is capable of moving across area and remotely guided by a person who is directed and navigated using remote camera and computer.. Mobile robots using Zigbee protocol for the purpose of navigation using personal computer, implemented with wireless vision system for remote monitoring and control. Its main feature is its use of the Zigbee protocol as the communication medium between the mobile robot and the PC controller. The robot can be monitored only by authorized persons who are previously present in the database for security reasons. For this we utilize the Face recognition technology. It is a system which can automatically identify and verify the individuals face. Thus face and emotion recognition offers one of the most natural and less obtrusive bio metric measures of identification
OpenMind / Person FollowingA real-time person following system using visual tracking and re-identification for robotic applications.
HFUT-YYH / VMC Based QMRMechanical structure and software and hardware control system design that features Parameter identification system based on reinforcement learning, and visual recognition system based on yolov5 model.
Toemazz / ProductionLineVisualInspectionAn automated production line visual inspection project for the identification of faults in Coca-Cola bottles leaving a production facility
ai4ce / SeeUnsafeIntegrate language and vision for traffic accident identification, reasoning, and visual grounding
utiasSTARS / Inertial Identification With Part SegmentationSource code implementing our visual part-segmentation pipeline and inertial identification algorithms, and our contributed dataset of objects used to test the whole pipeline.
HCIS-Lab / PF BCP[ICRA 2025] Potential Field as Scene Affordance for Behavior Change-Based Visual Risk Object Identification
