Object Detection / Scene Detection

Object and Scene Detection is a specific implementation of Visual-AI (also known as computer vision or vision-ai) that allows a wide range of common objects to be detected and tagged in visual media (images and videos), and where relevant, the context, or overall scene, can also be derived. This includes not just top-level objects, such as ‘Mammal’ to be detected, but even multi-layered specific sub-types of an object, e.g. ‘Mammal > K9 > Labrador’.

This is more widely known as ‘Image Tagging’ where annotations are provided in a tree structure for each image or frame of video.

VISUA’s Object and Scene technology delivers fully categorised and contextualised data thanks to our unique, massive, and deep hierarchical tagging library.

Object classifications are organised as a hierarchical tree of labels and highlights their parent types within a wider taxonomy of the object and scene. This gives greater flexibility to analyse data at macro or micro levels of granularity and can be customised to your application needs. E.g. Mammal > K9 > Sausage Dog, Labrador, etc.

Objects and Scenes recognition also leverages curated semantic metadata to define connections between Objects and Scenes which are semantically related. These connections are sorted from the most concrete to the most abstract and are available for each Object or Scene recognized.

Totally! VISUA’s Object and Scene technology delivers fully categorised and contextualised data thanks to our unique, massive, and deep hierarchical tagging library.

Detected objects are therefore organised as a hierarchical tree of labels and highlights their parent types within a wider taxonomy of the object and scene. This gives greater flexibility to analyse data at macro or micro levels of granularity and can be customised to your application needs. E.g. Mammal > K9 > Sausage Dog, Labrador, etc.

Objects and Scenes recognition also leverages curated semantic metadata to define connections between Objects and Scenes which are semantically related. These connections are sorted from the most concrete to the most abstract and are available for each Object or Scene recognized.

In contrast to other computer vision solutions that provide a one-size-fits-all offering, VISUA does not have a standard price list. This is for a very logical reason – There are many factors and combinations of settings that define the final cost for each customer and our Visual-AI (Computer Vision) solutions are very flexible in this regard so that the final implementation not only meets each customer’s specific technical needs, but also budgetary needs.

For instance, these are just some of the factors that influence the final cost:

• The volume of images/videos that you need processed
• The resolution of media to be processed
• The minimum size of logos to be detected in the media as a percentage of the overall frame
• For video, the frame rate you wish to process
• How much of the object/s must be visible (how occluded it may be) that you wish to detect
• If you are looking for general object detection or you wish to find only exact matches of a specific object
• If you want to combine technologies, such as object and scene with logo detection, so that you can annotate the objects where logos were found (often called placements), such as types of billboards, caps, shirts, bottles, etc.
• How quickly you want to receive the results, i.e. real-time vs next day

All these factors, plus some other more obscure ones allow us to optimise the offering to deliver the very best value for any use case and scale.

Absolutely, we actively encourage and are very happy for our customers to benchmark our Visual-AI (Computer Vision) tech stack against other providers as we typically out-perform them.

However, as we don’t provide a one-size-fits-all system, we like to discuss your specific use case and requirements. Based on the outcome of that discussion, we then set up a live test using your own data. Once complete you receive the results and annotations in whatever format you need and we are available to discuss the specifics with you.

This is completely free, so simply get in touch to set this up.

Our Visual-AI (Computer Vision) technologies and API focus on the processing of visual media for the purpose of detecting logos, objects and text within images or indeed visually similar copies of a source image. This is typically carried out for client/partner companies who already have access to their own source data.

For some specific projects we can, and have, assisted in the collection of data for processing. However, this is the exception and there are requirements, such as minimum volume and data licensing requirements.

If this is something you might require, please get in touch.

Our Visual-AI (Computer Vision) technologies and API focus on the processing of visual media to identify and report on key visual signals. The data is fed back to clients for them to make use of in their service or platform.

VISUA deliver data accuracy (precision and recall) in the range of 98.7% average precision and between 90-99% recall, (recall varies based on use-case). For more detail on Precision and Recall see the relevant question below. VISUA also uses humans to constantly sample-check and confirm the accuracy of the AI derived data, so data verification by the customer is rarely necessary.

Moderation of content is different. This is useful where the detection of an element (such as a brand) is correct, but the context of its use may be ambiguous. This is especially the case in use cases around copyright and trademark infringement or product counterfeits, where marginal cases need a human review for final decision.

In most cases, customers have their own Trust and Safety teams to review and moderate content, but for specific projects we can, and have, assisted in this task. However, this is the exception and there are requirements, such as minimum volume and data licensing requirements.

If this is something you might require, please get in touch.

The simple answer is no. We already process billions of images and millions of hours of video per month and have the ability to scale up for heavy demand at almost a moment’s notice.

Many world-leading companies already trust our technology to deliver high-volume processing for them, so if you need computer vision / Visual-AI at scale, you’ve come to the right place.

VISUA’s Visual-AI (computer vision) tech stack is built and optimised to handle massive volumes of data in the millions of media items per customer per month. Lower volumes can be supported, but typically, the lower limits are in the thousands of media files per day.

If your volume requirements are smaller than that then it may be worth reaching out to one of our customers in your specific sector, who will be able to support your needs better.

However, we do understand that some of the largest projects came from small beginnings. Also, some academic studies have relatively small processing requirements. So, if you have a new or academic project that you’d like to discuss, please do reach out. We’d be happy to discuss further.

VISUA’s object and scene library is one of the largest in the industry and constantly growing. It currently stands at over 11,000 discrete objects and scenes and therefore covers all major common objects, along with variants, to the highest levels of confidence.

Absolutely! New objects or highly distinct variants of objects can be added as required.

The process of adding new or distinct objects to the library is quite straightforward. The first step is to make contact with our team, from there a discussion will be organised to understand the nature of the object in question and its taxonomy. We will then determine the number of example images containing that object and will train the model for you.

The precise number of images and other requirements will depend on the uniqueness of the object in question. If it’s a minor variant of an existing object in the library, then a tweaking of the existing model may be sufficient, whereas for more obscure or unique products with generic IDs, it may require more in-depth training using our Custom Object Detection system.

The time required to train a new model varies depending on the object in question and use case. However, in most cases it takes a day or two to add a new object to the library.

Absolutely! Indeed some of our most interesting and unique applications have been on-device. Of course, every project is different and requirements vary, so if On-Device deployment is critical for your project, please do get in touch to discuss further.

Great question. This is another quite unique offering from VISUA. Deployment can be implemented in the cloud, on-premise, and even on-device. You can even choose a combination of all three if required.

Of course. For very unique objects we use our Custom Object Detection stack. This allows virtually any object to be trained and recognised in any form of visual media.

Yes. VISUA’s Object & Scene Detection API works with all popular image and video file formats, including streaming media.

VISUA’s Object & Scene Detection API supports all popular Images and video formats. This includes GIFs and even streaming video formats.

There is no specific minimum resolution as such, however, lower resolution media would also impact on the size and quality of objects contained in the media. This would therefore require specific tuning of the object and scene detection API in order to maximise the accuracy of detections.

With regard to maximum resolution, our resolution can process media files up to 4K resolution.

Yes. Our Object & Scene Detection API is very flexible and can be specifically tuned per customer use. We call this ‘Occlusion Tolerance’ and allows the level of sensitivity to obscured objects to be tuned in a range from zero occlusion (fully visible) to high occlusion (highly obscured or cropped).

It’s usually best to organise a short discussion to determine your specific requirements and from that a live test can be organised.

Yes. This is another tuning option because the smaller the object to be detected, the harder the Object Detection Visual-AI (computer vision) needs to work. However, this is also linked to the resolution of the media because the higher the resolution, the higher the quality of the object (more pixels), so it is less intensive to accurately detect smaller objects in a very high-quality image/video.

It’s usually best to organise a short discussion to determine your specific requirements and from that a live test can be organised.

Yes, of course. We call this ‘Matching Tolerance’ and it allows you to specify the percentage match that you wish to report on. For instance, you may only want to see detections that are a 100% match of your source object. However, in many cases, our clients prefer to include close matches that would include modified versions of the object.

It’s usually best to organise a short discussion to determine your specific requirements and from that a live test can be organised.

Yes, that is possible. However, we would use a different technology for this called Visual Search. This doesn’t specifically look for objects within media, but compares the overall image itself to see how much it matches the source image.

In other words, if you want to see where your product appears generally in media, then you should use the Object & Scene API. If, instead, you simply want to see where the image itself is used, then Visual Search is best for that.

We are proud to deliver the industry’s most accurate Visual-AI (computer vision) technology stack. This has been confirmed on many occasions where clients have tested numerous providers against our tech as part of their due diligence testing. In fact, we always encourage prospective clients to run tests against other solutions and compare the results with our Visual-AI.

The main reason for this is the flexibility our API provides and the unique ability to tune the stack to deliver the very best results for each use case.

In real terms VISUA delivers data accuracy (precision and recall) in the range of 98.7% average precision and between 90-99% recall, (recall varies based on use-case). For more detail on Precision and Recall see the relevant question in this FAQ.

Precision and recall are critical terms when it comes to Visual-AI (computer vision) and together equate to the overall accuracy of the detections. Each term relates to either false positives or false negatives as follows:

Precision = False Positives

a test result which wrongly indicates that a particular condition or attribute is present. In other words seeing something that is NOT actually there.

Recall = False Negatives

a test result which wrongly indicates that a particular condition or attribute is absent. In other words NOT seeing something that IS actually there.

Currently VISUA boasts 98.7% average precision and between 90-99% recall, (recall varies based on use-case).

If this is a key KPI for your use case then get in touch to organise a more in depth discussion and a live test.

This is another flexible option when using our Object & Scene Detection API. Our ‘Intelligence’ option allows you to choose whether to receive basic binary present/not present for your detections or advanced intelligence, such as size in frame, position in frame, time on screen (for video), etc.

Complete details are available in our API Documentation, but if you would like to discuss this further, please reach out and a call can be organised.

Detection and annotation data are typically provided in JSON, XML or CSV format. Please get in touch if you require an alternative format.

Our object & scene Visual-AI (computer vision) API can process data to virtually any schedule you require. This can include real-time processing (used most often by broadcast and sports sponsorship monitoring platforms) or as long as 24 to 48 hours (as often used by brand monitoring companies). The speed of processing is another factor in cost, so please discuss this with us further.

We like to think that our Visual-AI (computer vision) API is very easy to implement as part of any workflow, in fact, in most cases implementation takes as little as two hours. We have very clear API documentation also. But we are not simply an API provider, so do not hesitate to get in touch with any questions you may have. We also implement a very thorough onboarding process and as a client you will have direct access to our team for any ongoing support questions.

Yes, you can find very clear API documentation for our Object & Scene Detection endpoint, or indeed any of our other technologies. You can find all object & scene detection documentation here

Absolutely! Unlike other solutions on the market that charge significant fees for support, or force you to reach out to third-party consultants, VISUA is proud to be much more than simply an API provider. You can get in touch with any questions you may have during your research and feasibility stage. We also implement a very thorough onboarding process and as a client you will have direct access to our team for any ongoing support questions.

For sure! Many of our partner clients came to us with quite unique requirements. A short discussion will allow us to gather your requirements and determine how easily we might support it.

Every offering from each company has a slightly different focus. The differences are too numerous to outline in this FAQ. However, we have developed specific comparison documents, which are available in our blog. Specifically, you can find a Google Cloud Vision vs VISUA and Amazon Rekognition vs VISUA docs. More are being added regularly.

If you have specific questions, please don’t hesitate to get in touch.

For sure! You can combine object and scene detection with text detection and logo detection to begin to understand context and sentiment from visual media.

In fact not only is it technically possible, we have built our API to make this as simple as possible. Our ‘Batch Task Processing’ allows multiple tech stack requests to be made in a single call. See our API Documentation for more details.

Yes, we have specific commercial initiatives to support these types of projects, although there are some qualifying requirements. Please get in touch to see if your project qualifies for support.