Webinar follow-up: Q&A on IPRally’s Image Search
Last week's Image Search webinar, hosted by Sakari Arvela and myself, was a great success with nearly 300 attendees from 27 countries. We had an incredibly engaged audience who asked about 60 questions throughout the session. While we addressed many questions during the webinar, we couldn't get to all of them. So we have compiled this Q&A to share comprehensive answers with everyone. Thank you to all who joined and made the session so interactive!
Basic functionality
What exactly is IPRally's Image Search?
It's not a traditional image similarity search – rather, it converts images into descriptions which are then used to search patents. The system analyzes the content of your uploaded image to find patents with related concepts or technologies, regardless of visual similarity.
Do I need a special subscription to use image search?
No, image search is available to all IPRally users without additional cost.
What image formats are supported?
JPEG, PNG, and WEBP formats are supported, with a maximum file size of 20MB.
Can you upload multiple images for a single search?
Currently, only one image per search is supported. However, there are plans to implement multiple image support in the coming year, which would allow users to show different aspects of an invention (e.g., external appearance and internal structure) in a single search.
How does the system generate the search from an image?
The system uses Anthropic's Claude 3.5 Sonnet model to analyze the image and convert it into a textual description. This description is then transformed into a search graph, which is used to find relevant patent documents. When building our patent database, we process only the patent text into a searchable form - we don't include patent images in this process. This means your image search will find patents based on their technical content, not based on any images they contain.
In other words, while your search starts with an image, the results are found by matching the meaning of that image against the meaning of patent texts.
What do you mean by "graph" in this context?
When we talk about a "graph" in IPRally, we don't mean a chart or a visual diagram. Instead, we mean a structured representation of technical features and their relationships – like a family tree of the invention's components and how they connect to each other. For example, if the invention is a vehicle with wheels, the graph would show "vehicle" as a main component connected to "wheels" as a subcomponent. This structure helps our AI understand the technological relationships more accurately than plain text would.
Search capabilities
What types of images work best?
Most types of images work well, including:
- Hand-drawn sketches
- Technical drawings
- Photos
- Product sheets
- Chemical structures
- Images without labels
- Diagrams with reference numbers
Do images need to have labels or text?
No, the system can interpret images without labels. However, you can provide additional context in the text field below the image to clarify or add information not visible in the image.
How does it handle reference numbers in technical drawings?
If a drawing only shows reference numbers without labels, you can provide the meanings in the additional context field (e.g., "1: Rotor, 2: Stator, 3: Bearing").
Can it search for design patents?
No, currently the system only searches utility patents as design patents are not in the searchable database.
Search results
How should I interpret the AI score for image searches?
The AI score for image searches works the same way as in other IPRally search types. What constitutes a "good" score depends on several factors:
- Searches focusing on partial features typically get lower scores than searches for complete inventions
- Score interpretation varies by technical field
- Typically the threshold for good results is somewhere in the 40–70 range, depending on the search
- With experience, you'll develop a better understanding of typical scores in your domain
We recommend focusing not on the absolute value of the scores but on their relative differences. Start your analysis from the top and work downward as long as it feels useful. You can safely ignore the score if you don't find it useful.
Can I refine the search results?
Yes, you can:
- Edit the resulting graph directly
- Modify the context text
- Apply boolean filters
- Use the "Zoom to Favorites" feature to find more results similar to ones you've marked as relevant
When I edit the additional context field, should I keep the old text?
Each time you edit the additional context field and run a new search, the system uses only the current content of the field. This means if you want to keep certain search aspects that are working well, make sure to keep that text in the field. You can freely add new information or remove text that isn't producing useful results.
Language and international coverage
Does it work with non-English images (e.g., diagrams with German, Korean, or Chinese text)?
Yes, the system can interpret images containing text in any language.
How does it handle non-English patent documents in the search results?
When processing patent documents for its database, IPRally translates non-English documents into English before converting them to graph format. While search quality might be slightly lower for translated documents, the system can find patents in any language. All searches are performed against both native English documents and translated documents.
Security and Data Protection
How secure is the image search feature?
IPRally implements industry-standard security practices for image search:
Data Encryption and Storage:
- All data is encrypted during transfer using HTTPS
- Images and associated data are encrypted with customer-specific keys when stored in IPRally's database
- Your company retains ownership of all images and generated outputs
Data Processing:
- Images and additional context are processed securely within IPRally's systems
- Only the image and additional context are shared with Anthropic (our AI provider) – no customer or user identification is included
- Anthropic stores this data for 30 days to meet legal obligations
- IPRally stores your images securely so you can return to your searches later
Privacy:
- IPRally engineers cannot access your images or search data
- Images are not used for training AI models
- No customer-identifying information is shared with third parties
For detailed information about security controls and terms:
Can I use confidential images like invention disclosures?
Whether you can use confidential images in searches depends on your company's specific policies. If your company has approved IPRally's Generative AI Terms and is comfortable using our Ask AI feature, then Image Search should not pose any additional risks – we handle image data with the same security measures as Ask AI questions. However, we recommend checking your company's specific policies regarding sharing confidential information with AI tools.
Where can I watch the webinar recording?
The webinar is available on-demand here.
Can I try IPRally?
Yes, we’re happy to offer you 3 days of free access to IPRally, including Image Search. No costs, no commitments – just the opportunity to see how it can change the way you search.
