📷

Computer Vision Lab

Vertex AI Workbench with NVIDIA T4/A100

Overview

The Computer Vision Lab is where you train machines to see. It combines GPU compute for model training with a suite of annotation, visualization, and dataset management tools purpose-built for image and video data. Datasets are stored on Cloud Storage and streamed efficiently to your GPU for training — no downloading gigabytes of images to local disk. The lab comes pre-loaded with benchmark datasets (COCO, ImageNet subsets, custom annotated datasets) and state-of-the-art model implementations (YOLOv8, Detectron2, SAM). Label Studio and CVAT are available for hands-on annotation exercises.

What You'll Do in This Lab

  • Annotate images with bounding boxes, polygons, and segmentation masks in Label Studio
  • Train YOLO object detection models on your annotated datasets
  • Run Segment Anything (SAM) for zero-shot and prompted segmentation
  • Build real-time video processing pipelines with OpenCV and PyTorch
  • Apply image augmentation pipelines with Albumentations
  • Evaluate models with COCO metrics — mAP, IoU, per-class precision/recall

Lab Workflow

1

Launch

Start the lab. A T4 GPU instance with OpenCV, YOLO, and Detectron2 pre-installed launches. Datasets load from Cloud Storage.

2

Annotate

For annotation exercises, open Label Studio in your browser. Draw bounding boxes, segment objects, or classify images.

3

Train

Load annotated data into your training pipeline. Train YOLO, Mask R-CNN, or custom architectures with GPU acceleration.

4

Evaluate

Run COCO-style evaluation. Review per-class mAP, visualize predictions vs ground truth, and identify failure cases.

5

Visualize

Render detection/segmentation overlays on images and video. Build OpenCV visualization pipelines for demo outputs.

6

Export

Save trained models. Export for edge deployment exercises or push to model registry.

Hardware & Environment

Machine Type n1-standard-8 (8 vCPU, 30 GB RAM)
GPU 1x NVIDIA T4 (16 GB VRAM)
Datasets COCO (subset), ImageNet (subset), custom annotated sets on Cloud Storage
Annotation Tools Label Studio and CVAT on dedicated web endpoints
Video Processing OpenCV 4.x with CUDA-accelerated video decode
Session Length 2-4 hour sessions, annotations persist between sessions

Pre-installed Tools

OpenCV Ultralytics YOLOv8/v9 Detectron2 Segment Anything (SAM) Label Studio (annotation) CVAT

Frequently asked questions about this lab

What is the Computer Vision Lab? +
GPU-accelerated environment for image and video processing. Pre-loaded with large-scale vision datasets, annotation tools, and state-of-the-art model implementations.
Which courses use this lab? +
This lab is included in: Computer Vision & Visual AI, AI for Robotics & Edge Computing.
What hardware does this lab run on? +
Vertex AI Workbench with NVIDIA T4/A100. Machine Type: n1-standard-8 (8 vCPU, 30 GB RAM); GPU: 1x NVIDIA T4 (16 GB VRAM); Datasets: COCO (subset), ImageNet (subset), custom annotated sets on Cloud Storage; Annotation Tools: Label Studio and CVAT on dedicated web endpoints.
What software comes pre-installed? +
Comes pre-loaded with OpenCV, Ultralytics YOLOv8/v9, Detectron2, Segment Anything (SAM), Label Studio (annotation), CVAT. No local installs or dependency setup required — open your browser and start working.
Can I bring my own datasets and code into this lab? +
Yes. Datasets can be uploaded directly or synced from Google Cloud Storage. Notebooks and source files have built-in Git integration so you can push work to your own GitHub or GitLab repos.
Do I need to enroll in a course to use this lab? +
Yes. Lab environments are provisioned per-student as part of an AI Labs course enrollment. Browse the courses linked above to find programs that include this lab.

Related labs

Other AI Labs environments students typically use alongside this one.

Ready to Try This Lab?

Enroll in a course that uses this lab, or visit our Houston center for a hands-on demo.