Model reference — Video
Video OCR reads the text that appears in footage — signage, vehicle markings, container and wagon numbers, screen captures, on-screen graphics and burnt-in subtitles — and returns it with the time and position at which it appeared.
Two features distinguish it from reading a document. Text appears for a stretch of frames rather than once, so the same words must be recognised as one occurrence and the clearest frame used, which both improves accuracy and cuts cost enormously. And text in the world is at an angle, moving, partly obscured and unevenly lit, so results are much better when a detection model first locates the text region across frames and only that region is read.
Models in this group take a recorded video file as input and are applied frame by frame. Each specification table gives three hardware tiers — Minimum, the smallest setup on which the model runs correctly; Medium, the usual production configuration; and High, a configuration sized for peak volume. The rate is the number of sample inputs processed per hour on that hardware. Use these rates for initial sizing. Before production, benchmark your own data to validate accuracy, latency, throughput and cost. The sample input here is one minute of 1080p video at 25 frames per second.
Vendor: Baidu (PaddlePaddle)
What it does: reads text quickly on sampled frames, with good tolerance of angle and low resolution. The usual default for bulk video reading.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 6 GB | 24 GB | 80 GB |
| vCPUs | 4 | 8 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (video minutes processed per hour) | ≈ 300 | ≈ 1,200 | ≈ 3,400 |
Vendor: Jaided AI
What it does: covers eighty languages in a light model, useful as a first pass to find which stretches of footage contain text at all.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 6 GB | 24 GB | 80 GB |
| vCPUs | 4 | 8 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (video minutes processed per hour) | ≈ 400 | ≈ 1,600 | ≈ 4,500 |
Vendor: Datalab
What it does: reads text in roughly ninety languages and returns positions, which suits international signage and multilingual broadcast material.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (video minutes processed per hour) | ≈ 70 | ≈ 300 | ≈ 600 |
Vendor: Microsoft
What it does: reads difficult individual text regions — handwritten or stylised — once another model has located them.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (video minutes processed per hour) | ≈ 200 | ≈ 800 | ≈ 2,000 |
Vendor: StepFun
What it does: reads structured text such as on-screen tables and scoreboards, returning the layout rather than a stream of words.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 3060 | RTX 4090 | A100 80 GB |
| VRAM | 8 GB | 24 GB | 80 GB |
| vCPUs | 6 | 12 | 24 |
| RAM | 16 GB | 32 GB | 64 GB |
| Server | 1× RTX 3060 12 GB | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB |
| Rate (video minutes processed per hour) | ≈ 60 | ≈ 220 | ≈ 450 |
Vendor: Alibaba Cloud
What it does: reads text that requires context to interpret — a partially obscured sign, a number on a moving vehicle — using the surrounding scene. The most accurate and the slowest option.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | RTX 4090 (7B, reduced precision) | A100 80 GB (7B, full precision) | 2× H100 80 GB (72B model) |
| VRAM | 16 GB | 80 GB | 160 GB combined |
| vCPUs | 8 | 16 | 32 |
| RAM | 32 GB | 64 GB | 200 GB |
| Server | 1× RTX 4090 24 GB | 1× A100 SXM 80 GB | 2× H100 SXM 80 GB |
| Rate (video minutes processed per hour) | ≈ 40 | ≈ 160 | ≈ 90 (72B model, higher accuracy) |
Vendor: FFmpeg project
What it does: decodes the video and extracts the frames to be read, and is where a well-chosen sampling rate cuts the cost of the whole pipeline by an order of magnitude.
| Requirement | Minimum | Medium | High |
|---|---|---|---|
| GPU type | No GPU required | No GPU required | GPU-accelerated decode (NVENC/NVDEC) |
| VRAM | — | — | 8 GB |
| vCPUs | 2 | 8 | 16 |
| RAM | 4 GB | 16 GB | 32 GB |
| Server | CPU instance, 2 vCPU | CPU instance, 8 vCPU | 1× RTX 4090 24 GB |
| Rate (video minutes processed per hour) | ≈ 1,200 | ≈ 4,000 | ≈ 12,000 |
What is readable depends on resolution, motion and how long text stays on screen. Our consultants review your footage and the text you need captured, then recommend the frame-sampling strategy and the reading model, with measured accuracy on your own material.
At the start of a project we may run a short proof of concept on a sample of your own data, measuring the accuracy and the throughput the model actually achieves on your material. That replaces the estimates on this page with real figures, so the cost and the schedule for the full engagement are known before it is committed.
Share a representative sample, expected volume, latency target and deployment location for Video OCR / text recognition. We benchmark the shortlisted models, recommend the lowest-cost GPU configuration that meets the target, and scale it from pilot capacity to a dedicated production cluster.