Roboflow Reviews & Product Details

Easy to user interface, great free datasets, loads fast. Quite affordable. Customer support is bad, got no response on trying to upgrade to Enterprise - which has most of the needed features for teams. Using it for over a month. AI auto label works well, which is the main selling point i guess. Review collected by and hosted on G2.com.

The RAM it consumes, we loaded 20 000 images into labelling and it takes 2GB. Just one tab of their site. Looks like they preload all images into frontend.

There is no way to see what other labelers labelled. Requested twice to upgrade to Enterprise but got no response over time.

On object detection, it is way too easy to misspell the class name and by mistake create new class. There must be some way of validating or modal showing up to let the user know that they are creating new class. Thats during labelling. E.g. class name is ar_5 , i typed ar 5 and clicked enter by mistake - done, object got new class. Review collected by and hosted on G2.com.

User-friendly, fast, and with many options to use. In my case, it’s especially helpful for image labeling, and it works well in Google Colab. Review collected by and hosted on G2.com.

To be honest, nothing has really bothered me since 2022. I’ve been working with it since then, and the only thing is that when an update introduces a new feature, I sometimes need a bit of time to figure it out. But overall, it’s brilliant. Review collected by and hosted on G2.com.

Roboflow makes computer vision workflows simple and fast. Dataset management, labeling, augmentation, and versioning are all in one place, which saves a lot of engineering time and reduces mistakes. Review collected by and hosted on G2.com.

Some advanced features are locked behind higher pricing tiers, and large datasets can become expensive. For highly custom pipelines, there can be some limitations compared to fully self-hosted solutions.Some advanced features are locked behind higher pricing tiers, and large datasets can become expensive. For highly custom pipelines, there can be some limitations compared to fully self-hosted solutions. Review collected by and hosted on G2.com.

Roboflow distinguishes itself as a leading end-to-end computer vision platform by streamlining the often disjointed process of object detection and instance segmentation into a cohesive, unified workflow. One of the most notable benefits is the annotation experience, particularly the "Smart Polygon" feature powered by the Segment Anything Model (SAM). This tool transforms the usually painstaking, pixel-level clicking required for segmentation masks into a fast, one-click operation. In addition to this convenience, the platform provides significant strategic value through its powerful preprocessing and augmentation tools, which let you expand your dataset and enhance model robustness by simulating conditions like noise or rotation—without the need to gather new images. Coupled with "Label Assist" for automated pre-labeling and the capability to export datasets in more than 30 widely used formats (including YOLO, COCO, and TFRecord), Roboflow effectively eliminates the technical barriers between data preparation and training. This allows you to dedicate your efforts to refining model performance instead of dealing with file conversions. Review collected by and hosted on G2.com.

What I find frustrating about Roboflow is that its AI features act more like assistants than fully autonomous agents, especially regarding the demanding tasks of annotation and Quality Assurance (QA). When I upload a dataset, I expect the AI to handle image labeling from start to finish. However, I still have to go through each image myself to apply or confirm labels, which makes the process feel unnecessarily manual. This issue becomes even more pronounced with instance segmentation, where precise polygon annotations are essential. The platform does not provide an automated QA layer to guarantee the accuracy of these annotations, so I am left with the responsibility of checking every mask for correctness. I wish there were an AI 'supervisor' to automatically validate the quality and consistency of the data, rather than leaving all the verification work to me. Review collected by and hosted on G2.com.

I like that i can use credits to train quick custom models that speed up the labeling process big time. The provided tools are flexible enough to quickly label images or copy labels over to other images. Exporting training data is also very flexible allowing me to export it in every required format based on the needs of the project. it integrates well with our internal workflow. When i had difficulties, customer support reached out on their own to me since i was using a new feature and helped me continue. I use it at least once a week. Its easy to implement in every current project because of its flexibility. Review collected by and hosted on G2.com.

When importing yaml files it sometimes messes up the order of the classes, arranging them alphabetically. this requires one more step to bring them back to the original order. This could for sure be avoided. Review collected by and hosted on G2.com.

What I like best about Roboflow is how seamlessly it handles every step of dataset preparation — from uploading and annotating images to preprocessing, augmentation, and exporting for training. The platform is extremely user-friendly, and it saves an incredible amount of time when working with large datasets. The automated features, version control, and YOLO export options made it easy to manage our BharatPotHole dataset of 7,000+ frames efficiently. For our iWatchRoad pothole detection research, Roboflow helped maintain consistent labeling quality, supported collaboration, and simplified the entire workflow. It’s one of the few tools that genuinely makes computer vision dataset management effortless. Review collected by and hosted on G2.com.

Honestly, there’s not much to dislike. Occasionally, large uploads can take a bit longer, especially when dealing with thousands of high-resolution images, but that’s expected with big datasets. Other than that, everything—from annotation tools to export options—has worked smoothly. Review collected by and hosted on G2.com.

Quick to implement on NVIDIA Jetson device, easy to train models, lot's of different training options, workflows with tons of tools to automate processes. Support team is quick to respond and help answer questions and resolve issues. Constantly innovating. Able to easily integrate with other tools we use. We use it every day for photo and video analysis and it is able to keep up as we scale. Review collected by and hosted on G2.com.

Wish it was easier to view and manage live stream video inference with their Growth subscription. Review collected by and hosted on G2.com.

Roboflow has become a key tool for computer vision research, especially in projects focused on real-time detection. What I like is its focus on simplifying the entire cycle for training models with datasets or your own datasets, which allows you to focus on data quality and algorithm optimization instead of wasting time on technical configurations. Review collected by and hosted on G2.com.

While Roboflow is a leading tool, the least useful feature is the integrated training tool for expert users. Auto-labeling requires a lot of human supervision (human-in-the-loop), and for very small objects or low-resolution images, Roboflow's AI often fails, requiring exhaustive manual labeling. Review collected by and hosted on G2.com.

As a researcher working on instance segmentation for medical imaging, I found the platform exceptionally helpful in streamlining the entire CV pipeline. The annotation tools are intuitive and precise polygon and smart polygon features saved significant time on detailed dental structures without sacrificing accuracy. Dataset versioning, augmentations (especially relevant ones like brightness/contrast for X-ray-like images), and seamless export to formats like COCO/YOLO made experimentation much faster and reproducible. Review collected by and hosted on G2.com.

The UI is sometimes not responsive, while switching workspace. Review collected by and hosted on G2.com.

Roboflow allows data annotation, and the results can be easily exported to any of the SOTA object detection frameworks. The data augmentation features make it fantastic—there’s no need to write code or spend additional time augmenting the dataset. In addition, every edit made to the dataset creates a newer version, which helps me track progress and monitor version control for the dataset. Review collected by and hosted on G2.com.

Not really a downside. If anything, it’s the limited availability of units for model training. Of course, we all know GPUs aren’t free… ahahhah. Review collected by and hosted on G2.com.