Show HN: Benchmarking VLMs vs. Traditional OCR https://ift.tt/daxherc

Show HN: Benchmarking VLMs vs. Traditional OCR Vision models have been gaining popularity as a replacement for traditional OCR. Especially with Gemini 2.0 becoming cost competitive with the cloud platforms. We've been continuously evaluating different models since we released the Zerox package last year ( https://ift.tt/5VIPSdG ). And we wanted to put some numbers behind it. So we’re open sourcing our internal OCR benchmark + evaluation datasets. Full writeup + data explorer here: https://ift.tt/U9sxq40 Github: https://ift.tt/jmMLbsi Huggingface: https://ift.tt/vuzDAZ1 Couple notes on the methodology: 1. We are using JSON accuracy as our primary metric. The end goal is to evaluate how well each OCR provider can prepare the data for LLM ingestion. 2. This methodology differs from a lot of OCR benchmarks, because it doesn't rely on text similarity. We believe text similarity measurements are heavily biased towards the exact layout of the ground truth text, and penalize correct OCR that has slight layout differences. 3. Every document goes Image => OCR => Predicted JSON. And we compare the predicted JSON against the annotated ground truth JSON. The VLMs are capable of Image => JSON directly, we are primarily trying to measure OCR accuracy here. Planning to release a separate report on direct JSON accuracy next week. This is a continuous work in progress! There are at least 10 additional providers we plan to add to the list. The next big roadmap items are: - Comparing OCR vs. direct extraction. Early results here show a slight accuracy improvement, but it’s highly variable on page length. - A multilingual comparison. Right now the evaluation data is english only. - A breakdown of the data by type (best model for handwriting, tables, charts, photos, etc.) https://ift.tt/U9sxq40 February 21, 2025 at 12:19AM

Show HN: WinCse – Integrating AWS S3 with Windows Explorer https://ift.tt/MyBRqwu

Show HN: WinCse – Integrating AWS S3 with Windows Explorer WinCse is an application that integrates AWS S3 buckets with Windows Explorer. Utilizing WinFsp and the AWS SDK, WinCse allows you to treat S3 buckets as part of your local file system, making file management simpler. The application is currently in development, with plans for additional features and improvements. https://ift.tt/F1emcZC February 20, 2025 at 11:23PM

Show HN: Make your logo liquid metal (open source) https://ift.tt/EU0s5XJ

Show HN: Make your logo liquid metal (open source) Good morning!! We thought the Apple liquid metal invite was so cool. How fun would it be if everyone could see their logo in liquid? So we made an app to let you make your logo in liquid. Just drag in your logo and see. To play with your logo: https://ift.tt/TeazPYR Repo: https://ift.tt/SHhNk6v (We think you're gonna love it!) https://ift.tt/TeazPYR February 20, 2025 at 01:41AM

Show HN: A new fork of OpenDeepResearcher with DeepSeek R1 https://ift.tt/8Hjr5bu

Show HN: A new fork of OpenDeepResearcher with DeepSeek R1 https://www.youtube.com/watch?v=aEOu9P4_2cU February 20, 2025 at 01:15AM

Show HN: Subtrace – Wireshark for Docker Containers https://ift.tt/hVSHQ07

Show HN: Subtrace – Wireshark for Docker Containers Hey HN, we built Subtrace ( https://subtrace.dev ) to let you see all incoming and outgoing requests in your backend server—like Wireshark, but for Docker containers. It comes with a Chrome DevTools-like interface. Check out this video: https://www.youtube.com/watch?v=OsGa6ZwVxdA , and see our docs for examples: https://ift.tt/XLlof79 . Subtrace lets you see every request with full payload, headers, status code, and latency details. Tools like Sentry and OpenTelemetry often leave out these crucial details, making prod debugging slow and annoying. Most of the time, all I want to see are the headers and JSON payload of real backend requests, but it's impossible to do that in today's tools without excessive logging, which just makes everything slower and more annoying. Subtrace shows you every backend request flowing through your system. You can use simple filters to search for the requests you care about and inspect their details. Internally, Subtrace intercepts all network-related Linux syscalls using Seccomp BPF so that it can act as a proxy for all incoming and outgoing TCP connections. It then parses HTTP requests out of the proxied TCP stream and sends them to the browser over WebSocket. The Chrome DevTools Network tab is already ubiquitous for viewing HTTP requests in the frontend, so we repurposed it to work in the browser like any other app (we were surprised that it's just a bunch of TypeScript). Setup is just one command for any Linux program written in any language. You can use Subtrace by adding a `subtrace run` prefix to your backend server startup command. No signup required. Try for yourself: https://ift.tt/XLlof79 https://ift.tt/VgWMDct February 19, 2025 at 04:59AM

Show HN: Streamer Emails – Quickly Find and Filter Twitch Streamers Emails https://ift.tt/nbfX1Vo

Show HN: Streamer Emails – Quickly Find and Filter Twitch Streamers Emails https://ift.tt/QTWsKmP February 19, 2025 at 03:33AM

Show HN: A GPU-accelerated binary vector index https://ift.tt/kO3vwMQ

Show HN: A GPU-accelerated binary vector index This is a vector index I built that supports insertion and k-nearest neighbors (k-NN) querying, optimized for GPUs. It operates entirely in CUDA and can process queries on half a billion vectors in under 200 milliseconds. The codebase is structured as a standalone library with an HTTP API for remote access. It’s intended for high-performance search tasks—think similarity search, AI model retrieval, or reinforcement learning replay buffers. The codebase is located at https://ift.tt/6dHvjpx . https://ift.tt/IhySljR February 17, 2025 at 06:15AM