计算机视觉 Jobs

Tengo un sistema de automatización ya operativo y necesito a alguien que lo lleve al siguiente nivel. La base está construida en Python y OpenCV, comunicándose con un ESP32; todo corre sin errores, pero falta completar la lógica de detección de objetos y la contabilización de productos. Lo que requiero: • Ajustar y optimizar el modelo de detección de productos (actualmente uso OpenCV + Python). • Implementar el conteo automático de cada producto que aparezca en la cámara. • Enviar el resultado al ESP32 a través de la interfaz existente para que el microcontrolador lo procese. • Dejar el código limpio, documentado y con instrucciones rápidas para volver a entrenar o cambiar el ...

# Micro1 Generalist Project — Looking for an Experienced Freelancer I’m looking for a freelancer who has experience with **AI generalist work, data annotation/data labeling, video analysis, and following detailed guidelines** to help me with a Micro1 Generalist project. ### Skills Required * AI/ML generalist knowledge * Video annotation and video analysis * Data labeling/annotation * Strong attention to detail * Ability to identify errors and inconsistencies in videos * Understanding and following detailed instructions and guidelines * Good written English and communication * Ability to work independently * Comfortable using online annotation/AI tools * Good time management * Ability to maintain consistent quality across multiple tasks ### What You Will Do The work may inv...

We're building a system that works similarly to sports auto-tracking cameras (e.g. Veo), but with a simpler, classical computer-vision approach — no deep learning or trained models needed. What we need: Camera stitching: Combine footage from two fixed cameras (mounted on one rig, overlapping field of view) into a single panoramic image (~180°), using calibration/homography. The cameras don't move relative to each other, so this should be a one-time calibration applied per frame. Motion-density tracking: From the panoramic feed, detect where players are concentrated (background subtraction / foreground blob density, not per-object classification) and use that to drive an automatic pan/crop — i.e. a virtual camera that follows the action without a human operator. ...

We're building a system that works similarly to sports auto-tracking cameras (e.g. Veo), but with a simpler, classical computer-vision approach — no deep learning or trained models needed. What we need: Camera stitching: Combine footage from two fixed cameras (mounted on one rig, overlapping field of view) into a single panoramic image (~180°), using calibration/homography. The cameras don't move relative to each other, so this should be a one-time calibration applied per frame. Motion-density tracking: From the panoramic feed, detect where players are concentrated (background subtraction / foreground blob density, not per-object classification) and use that to drive an automatic pan/crop — i.e. a virtual camera that follows the action without a human operator. ...

We're building a system that works similarly to sports auto-tracking cameras (e.g. Veo), but with a simpler, classical computer-vision approach — no deep learning or trained models needed. What we need: Camera stitching: Combine footage from two fixed cameras (mounted on one rig, overlapping field of view) into a single panoramic image (~180°), using calibration/homography. The cameras don't move relative to each other, so this should be a one-time calibration applied per frame. Motion-density tracking: From the panoramic feed, detect where players are concentrated (background subtraction / foreground blob density, not per-object classification) and use that to drive an automatic pan/crop — i.e. a virtual camera that follows the action without a human operator. ...

We're building a system that works similarly to sports auto-tracking cameras (e.g. Veo), but with a simpler, classical computer-vision approach — no deep learning or trained models needed. What we need: Camera stitching: Combine footage from two fixed cameras (mounted on one rig, overlapping field of view) into a single panoramic image (~180°), using calibration/homography. The cameras don't move relative to each other, so this should be a one-time calibration applied per frame. Motion-density tracking: From the panoramic feed, detect where players are concentrated (background subtraction / foreground blob density, not per-object classification) and use that to drive an automatic pan/crop — i.e. a virtual camera that follows the action without a human operator. ...

Urgent: Smartphone Video Data Collection Project (Multiple Slots Open) PROJECT OVERVIEW: We are urgently looking for remote participants to complete a quick, one-time face motion video collection project to train computer vision models. Multiple slots are open immediately. WHO CAN PARTICIPATE: • Male Candidates: Aged 36–50 • Female Candidates: Any age (18+) • Must have a valid smartphone (with front and back camera capability). • Must submit a basic identity verification document (School ID, Aadhaar, PAN, or DL) where private ID numbers are crossed out/hidden, but Name, DOB, Photo, and Country are clearly visible. PROJECT TASKS & GUIDELINES: This project includes different video tasks. Complete guidelines and a step-by-step training video will be shared vi...

I’m building out a series of AI-driven features and need an engineer who can step in wherever the development cycle demands— from early-stage prototyping to production deployment. The exact focus is still open, so you’ll help me evaluate whether classic machine learning, natural language processing, or computer vision best fits each feature we prioritise. Here’s how I see the collaboration: • Work with me to clarify the first use-case, audit the data I already have, and outline any additional collection or labelling steps. • Design a model architecture suited to that problem, train and validate it, then document performance clearly enough for non-technical stakeholders. • Package the solution into clean, modular code (Python preferred) with stra...

Project Description: Developed an end-to-end AI-powered medical intelligence platform for analyzing chest X-ray images and assisting with disease prediction. The system uses Deep Learning with DenseNet121 to classify X-ray images into five categories: Normal, Pneumonia, Atelectasis, Cardiomegaly, and Effusion. Integrated Grad-CAM Explainable AI (XAI) to generate visual heatmaps highlighting regions of the X-ray that contributed to the model's prediction. Built REST APIs using FastAPI for image upload, prediction, model information, and health monitoring. The platform also supports prediction history and is designed for integration with AI-assisted medical report generation using LLMs. Key Features 1. Medical Image Analysis – Processes chest X-ray images using deep learning. 2. ...

I’m looking for a Python-based workflow that takes my equirectangular photo collection and, for any two images I select, confirms whether they were shot from the same angle. Beyond the yes/no decision, the script must also: - check whether they are connected • calculate the scale ratio between the pair, - the angle yaw and pitch they connected • assign a reliability/confidence score to its assessment. sample dataset : i run the progam using cli, json output is fine All results should be written to a concise text report that I can easily parse or forward—feel free to suggest the most convenient plain-text structure. You’re free to use OpenCV, scikit-image, NumPy, or any other well-supported libraries so long as installation remains straightforward (p...

Remote Sensing + AI/ML Model for Identification of 10 Major NTFP Species in Jharkhand, India We are looking for an experienced "Remote Sensing / Computer Vision / Geospatial AI developer" to develop or fine-tune an AI/ML model capable of identifying and mapping "10 major Non-Timber Forest Product (NTFP) tree species in Jharkhand, India" using remote sensing imagery. Target Species 1. Sal 2. Mahua 3. Kusum 4. Tamarind 5. Kendu 6. Palash 7. Chironji 8. Amla 9. Harra 10. Bahera Objective The objective is to develop a reliable workflow that can identify these species from remote sensing data and ultimately generate a **species-wise tree inventory/map**, including tree locations and counts. Scope of Work We are open to either: Developing a new model from scratch or F...

I'm looking for a senior developer with hands-on experience defeating Arkose Labs FunCaptcha (specifically the newer game variants such as hopscotch_highsec / hopscotchv3) in a headless / API-based pipeline. ▎ ▎ I already have a working Node.js codebase that: ▎ - Mints Arkose tokens via a real-Chrome TLS client (bogdanfinn) ▎ - Rotates residential proxies + BDA fingerprints per attempt ▎ - Uses commercial classifiers (YesCaptcha, OmoCaptcha, ) ▎ ▎ The pipeline reaches the challenge-submit step consistently, but the classifiers we use are giving low-accuracy answers on the newer navigation-style variants, so we're getting solved:false on the final wave. ▎ ▎ What I need help with: ▎ - Diagnosing whether the failure is truly answer-quality vs a deeper trust/suppression issue ▎ - Imp...

I need a fresh collection of face video from Indian and Indonesian participants to expand a training dataset for an AI-driven facial recognition model. The emphasis is on video variety (angles, lighting, backgrounds, expressions) so the algorithm learns to generalise accurately across real-world conditions. Every image must be accompanied by three demographic labels: • Age range 18- 99 years • Gender : Male & Female Skin Tone: Dark for all age groups Medium for 35+ only Deliverables (per participant) Need to upload the video in our Client portal. Technical notes FPS 30 and Minimum resolution 720 px. Mobile: Vertical Recording Laptop: Horizontal Recording I’m working on a tight timeline, so please mention how many distinct contributors you can source an...

We need an experienced WebAR / computer vision developer to create a browser-based virtual piercing try-on. Users should open a webpage on iPhone/Android, activate the camera, and see a virtual stud realistically positioned on their ear. The key challenge is reliable ear/eartlobe tracking while the head moves. This is NOT a request to copy existing commercial software. We want an independently developed solution using open-source or approved licensed technology. FIRST MILESTONE – PAID PROOF OF CONCEPT Deliver a working browser demo that: * Uses live mobile camera * Detects/tracks the head and ear * Places a virtual 2–4 mm stud on the earlobe * Keeps the stud reasonably anchored during head movement/rotation * Works on iPhone Safari and Android Chrome Possible stack: Media...

I’m looking for a skilled mobile developer who can deliver a polished face-swap application that runs natively on both iOS and Android. The app must handle two core scenarios: • Real-time face swapping through the camera preview • Face swapping on photos selected from the user’s gallery After a swap, users need a smooth way to share the resulting image or short video straight to their favourite social platforms via the standard system share sheets or integrated APIs. Performance and believability are key; I expect fast, well-aligned swaps that hold up under normal lighting and movement. You’re free to choose the toolkit—OpenCV, MediaPipe, ARKit/ARCore, or another reliable library—as long as the final builds pass store review and run well on cu...

I have hours of CCTV footage from 5-a-side and 7-a-side turf matches and I want an end-to-end computer-vision pipeline that turns every recording into clear, per-player metrics. The system must treat fitness and football performance with equal weight. For fitness I care most about minutes played, distance covered and calories burned; for football performance the priorities are goals and shots. Anything else you can derive is welcome, but these numbers must be rock-solid. What I expect you to build • A repeatable pipeline that ingests raw CCTV video, detects and re-identifies each player, tracks their movement through the entire match and exports a CSV/JSON report plus simple visual overlays. • Fitness layer: automatic calculation of minutes on pitch, total distance, speed p...

I need a reliable hand to draw clean, tight bounding boxes on a set of images so I can feed them straight into an object-detection pipeline. Every occurrence of the target object in each frame must be boxed precisely—no clipped edges or excessive margins—so the model learns from accurate ground-truth data. You are free to work in any mainstream annotation tool you prefer (LabelImg, CVAT, Labelbox, VGG, makes no difference to me) as long as the final export matches the format I’ll specify before we kick off—COCO JSON or YOLO-style TXT are both fine. Please keep category labels consistent across the entire batch. Deliverables • Fully annotated image set with bounding boxes around every instance of the specified object(s) • Corresponding annotation file...

I'm looking to create engaging skits using computer vision technology for entertainment purposes. Key Requirements: - Develop creative and entertaining skit concepts leveraging computer vision. - Implement computer vision techniques like object detection or image classification. - Collaborate on scripting and staging for the skits. Ideal Skills: - Strong background in computer vision. - Creative thinking and experience in entertainment production. - Good scripting and storytelling skills.

My 3D scanner streams PCD point-cloud files that contain spiral-shank bolts mixed with other hardware. I need a compact, CUDA-friendly C++ solution that runs directly on an NVIDIA Jetson Orin NX, spots every bolt in each cloud, records the head-center XYZ (and, if practical, its axis direction), and writes the results to a structured XML file. Performance targets • Accuracy: at least 95 % correct identification on the annotated dataset I will supply. • Speed: real-time or near real-time processing on the Orin NX under the standard JetPack image. You are welcome to build on open-source libraries such as PCL, Open3D, Eigen, cuBLAS or TensorRT, provided everything can be compiled through CMake and redistributed without license issues. Deliverables • Well-documented...

We are looking for experienced Robotics Video Annotation Specialists to support an ongoing video annotation project. Prior experience with robotics data annotation is required. Experience working on Annotask or similar annotation platforms is strongly preferred, as familiarity with annotation workflows and task-based interfaces will be helpful. What you will do You will review robotics-related video data and complete annotation tasks based on detailed project guidelines. Accuracy and strict adherence to the provided rules are important. The initial requirement is to complete 2 tasks within 6 hours after receiving access and the assigned tasks. Qualification process Before starting paid tasks, selected candidates will receive the project guidelines and will need to complete a short qu...

I’m building a dataset to train computer-vision models, and I need recordings from 100 native residents of India performing simple head gestures. Each participant will record a short video in which they: • Nod up-and-down • Shake side-to-side A neutral background, steady lighting and the phone held at eye level are all that’s required; no special equipment beyond a standard smartphone camera is necessary. The clip can be silent, rotated in portrait or landscape, and should last roughly 15–20 seconds so both motions are clearly visible. Deliverables for you: 1. One MP4 or MOV video per participant labelled with a unique ID. 2. A brief text file (CSV or XLSX) mapping each ID to age range and gender for diversity tracking. 3. Written confirmation that ...

I’m building a browser-accessible tool that can take an image uploaded by the user, run it through an AI model, and immediately return classification or detection results on-screen. All core logic must live server-side, exposed through a clean REST or GraphQL endpoint, so the front-end remains lightweight and responsive across modern web browsers. Key expectations • Model accuracy matters: please start with a proven open-source architecture (e.g., YOLOv8, ResNet, EfficientDet) fine-tuned on a small sample set I’ll provide, then document how to retrain it when new data arrives. • One-click deploy: include a Dockerfile and concise README so I can spin everything up on a fresh VPS. • Results returned as JSON plus visual overlays (bounding boxes or masks) rend...