Computer Vision Tools

Forrester renamed this category between editions, from computer vision platforms to computer vision tools, and the four and a half years between the two evaluations contained the entire generative AI revolution. The first evaluation was a New Wave in November 2019, published before foundation models existed as a product category. The second is The Forrester Wave: Computer Vision Tools, Q1 2024, and the March 2024 scorecard is still the newest one under this name.

The gap contained the entire generative AI revolution. The scorecard that returned was written for the market that revolution produced.

The old scorecard, before the revolution

The Forrester New Wave: Computer Vision Platforms, Q4 2019, published November 26, 2019, scored eleven vendors on ten criteria: Amazon Web Services, Chooch AI, Clarifai, Deepomatic, EdgeVerve, Google, Hive, IBM, Microsoft, Neurala, and SAS.

Google led in a way that now reads like an artifact. It was the only provider to receive the highest possible differentiated score across all ten criteria, and it took the top Current Offering position outright. Clarifai was the other Leader, the only startup to earn a differentiated rating, on five of ten criteria.

The 2019 scorecard measured a market where the buyer's job was to train a model on their own labeled data. Data, capabilities, pre-trained models, development, deployment: the criteria walk through the full lifecycle of building vision from scratch. Strong Performers IBM, Deepomatic, and Hive and Contenders SAS and EdgeVerve filled out a field of platform builders.

Then came the gap. Foundation models arrived, and the question stopped being "what can you train" and became "what can you do on top of what is already trained." When Forrester returned to the market in March 2024, the category had been renamed, and the rename records what happened to the buyer. A platform is what you buy when the category is young. A tool is what you buy when you already own the rest.

The scorecard that returned: The Forrester Wave: Computer Vision Tools, Q1 2024

Principal analyst Indranil Bandyopadhyay announced the new evaluation on March 19, 2024 in a blog post framed around two trends: horizontal capabilities that promise breadth and vertical ones that ensure depth. His phrase for what buyers actually need: "building the bridges between pixels and purpose."

The Wave itself scores twelve providers on 28 criteria, and the criteria list is the new era in miniature. Data management, model development, data and model testing, model inference, architecture, complementary functionalities, and enablement. Training your own model is now one criterion cluster among several. Inference, orchestration, and enablement are the new load-bearing walls.

No newer edition has surfaced since March 2024, which makes this scorecard, published the same month as the first vendor announcements, the current map of the category.

Twenty three of twenty eight

Clarifai announced its Leader placement on March 28, 2024, and the number attached to it is the whole finding: the highest possible scores in 23 of 28 criteria.

Read the citation and the shape of the win appears. Forrester called Clarifai a "one-stop-shop CV platform" with a "quick-to-implement-AI mantra." The scored assets are not a proprietary model trained in-house. They are full-stack infrastructure, foundational models, model orchestration, fine-tuning, and supervision, under one umbrella. Forrester positioned it for public and private sector organizations in retail, manufacturing, finance, and media that want to build computer vision applications rapidly and securely.

Twenty three of twenty eight is not a strong showing. It is a different product. The 2019 field was scored on training your own models. The 2024 field is scored on orchestrating somebody else's, and the vendor that won did exactly that.

The pricing criterion in a tools Wave

H2O.ai took a Leader placement in its first year in the Wave, with the highest possible scores in twelve criteria including data and model testing, innovation, partner ecosystem, adoption, and one criterion that would have looked strange on the 2019 list: pricing flexibility and transparency.

A vendor best known for open-source machine learning landing a Leader spot partly on pricing is a signal, not a coincidence. When the models underneath are commodity foundation models, the differentiators move to the open layer: what the tool costs, how flexibly it licenses, and whether the stack you assemble can change pieces without changing invoices.

The data half of computer vision never left the market. It just turned into a criterion.

The annotation vendor in the tools Wave

The third publicly confirmed placement is Cigniti, positioned as a Challenger. Cigniti is an annotation firm, and its inclusion is the most instructive roster decision in the report. Its Zastra platform plans to apply active learning and generative AI to data annotation, with a roadmap covering lidar point data, synthetic data creation, and active learning audio annotation. Forrester's verdict: "a good match for clients seeking a data annotation accelerator" with platform capabilities around it.

A tools Wave with a labeling vendor in the field is Forrester stating the obvious thing every machine learning practitioner already knew: the models got easier and the data did not. The bottleneck moved upstream, and the scorecard followed it.

The scorecard predicted the exit

In May 2026, Nebius welcomed Clarifai's core team, including founder Matthew Zeiler, and licensed Clarifai's inference and compute orchestration technology while acquiring its patent portfolio. One exclusion in the deal says more than the inclusions: the license expressly leaves out Clarifai's legacy computer vision models.

The Wave's dominant vendor exited through its most scored criterion. Forrester scored the bridge in 2024. In 2026, the market bought the bridge, and the models stayed behind on purpose.

For a buyer, the lesson is concrete. The durable asset in this category is not the model. It is the orchestration layer around it, the same layer the 23-of-28 citation was built on. Any vendor pitch that leads with a proprietary model deserves a question about what remains when that model is replaced.

What to ask before you assemble the stack

Are you buying a platform or a tool? The rename is the first buying decision. If the stack around the vision layer already exists, a tool that plugs in beats a platform that replaces, and the scorecard does not make that call for you.

Who labels the data? The Challenger in this Wave is an annotation firm. Ask any shortlisted vendor how your data gets labeled, tested, and retrained, because the data half decides outcomes no matter how good the models are.

What survives a model swap? The Leader's own exit answers the question. Ask what you keep when the vendor's model strategy changes, and get the answer in the contract, not the demo.

Forrester does not endorse any vendor named here, and tier placement should not be read as a recommendation to buy.

Buyers evaluating this should also weigh AI Platforms, the closest coverage already published on this site: four years ago Forrester's AI platform evaluation was full of data science workbenches. The current one includes a CRM vendor, an RPA vendor, and an ITSM vendor, because agentic AI redrew what the category means.

The foundation layer underneath these vision models is scored separately in AI Foundation Models: in 2024 Forrester ranked the large language models themselves. Two years later the category split in two, and the split says something real about how enterprises actually buy AI.