fzysingularity

6 ms·

31.
by fzysingularity 8mo ago
ELO scores for OCR don't really make much sense - it's trying to reduce accuracy to a single voting score without any real quality-control on the reviewer/judge. I think a more accurate reflection of the current state of comp
32.
by fzysingularity 8mo ago
Apple OCR even on the Mac is insanely good, in fact way better than AWS textract/GCP cloud vision OCR. Any idea what model is being used?
33.
by fzysingularity 8mo ago
VLM Run ( https://vlm.run ) | Infrastructure Engineer + DevRel + AI/ML Engineer | Santa Clara, CA (HQ) VLM Run is building infrastructure for production Vision-Language Model (VLM) systems — fast inference, tool-use + orchest
34.
DeepSeek OCR 2: Visual Causal Flow(huggingface.co)
2 points by fzysingularity 8mo ago|0 comments
35.
by fzysingularity 9mo ago
> It's like going to the grocery store and buying tabloids, pretending they're scientific journals. This is pure gold. I've always found this approach of evals on a moving-target via consensus broken.
36.
by fzysingularity 9mo ago
I'd love to see Claude Code remove more lines than it added TBH. There's a ton of cruft in code that humans are less inclined to remove because it just works, but imagine having LLM doing the clean up work instead of the generatio
37.
Unified Vision-Language Agents – Detect, Segment, OCR, Generate and More(github.com)
5 points by fzysingularity 10mo ago|1 comments
38.
by fzysingularity 10mo ago
Here's a short cookbook exploring an agentic approach to vision–language tasks: detection, segmentation, OCR, generation, and combining classical CV tools with VLM reasoning. Happy to run examples if you leave a comment. [1] IPython no
39.
by fzysingularity 10mo ago
What is photopea built on?
40.
by fzysingularity 10mo ago
This is why arenas are generally a bad idea for assessing correctness in visual tasks.
41.
by fzysingularity 11mo ago
FYI one of the models on the battle was pretty slow to load. Are these also being rated on latency or just quality?
42.
by fzysingularity 11mo ago
I do think that if the agents can successfully resolve these tasks in a code execution environment, it can likely come up with better parametrized solutions with structured I/O - assuming these are workflows we want to run over and ove
43.
by fzysingularity 11mo ago
I definitely see the value and versatility of Claude Skills (over what MCP is today), but I find the sandboxed execution to be painfully inefficient. Even if we expect the LLMs to fully resolve the task, it'll heavily rely on I/O
44.
by fzysingularity 11mo ago
Claude does image generation in surprising ways - we did a small evaluation [1] of different frontier models for image generation and understanding, and Claude is by far the most surprising in results. [1] https://chat.vlm.run&#x
45.
VLM Showdown: GPT vs. Gemini vs. Claude vs. Orion(chat.vlm.run)
15 points by fzysingularity 11mo ago|1 comments
46.
by fzysingularity 11mo ago
We ran a small visual benchmark [1] of GPT, Gemini, Claude, and our new visual agent Orion [2] on a handful of visual tasks: object detection, segmentation, OCR, image/video generation, and multi-step visual reasoning. The surprising p
47.
by fzysingularity 11mo ago
Nice, that's pretty neat.
48.
by fzysingularity 11mo ago
SAM3 is cool - you can already do this more interactively on chat.vlm.run [1], and do much more. It's built on our new Orion [2] model; we've been able to integrate with SAM and several other computer-vision models in a truly comp
49.
by fzysingularity 11mo ago
SAM3 is cool - you can already do this more interactively on chat.vlm.run [1], and do much more. It's built on our new Orion [2] model; we've been able to integrate with SAM and several other computer-vision models in a truly comp
50.
by fzysingularity 11mo ago
Do you mean like creating a personalized item from another product image?
51.
by fzysingularity 11mo ago
Hey, thanks! Curious what you tried to test it. Segmentation models like SAM2 only gets you so far, but by make this instruction-driven with reasoning in the loop, it's remarkable what you can do these days. Stay tuned for more updates
52.
by fzysingularity 11mo ago
It's still early days, but we'll expand to more capabilities very quickly given that we're not bottlenecked by training a single large VLM to do these tasks - think video tracking, in-image editing, and 3D.
53.
Show HN: Chat with Orion – a visual agent that sees, reasons and acts(chat.vlm.run)
22 points by fzysingularity 11mo ago|10 comments
54.
ChatGPT uses YOLOv8 to detect UI elements(twitter.com)
1 points by fzysingularity 11mo ago|0 comments
55.
by fzysingularity 1y ago
If I had to guess, the OpenAI open-source model got delayed because Kimi K2 stole their thunder and beat their numbers.
56.
by fzysingularity 1y ago
If I had to guess, the OpenAI open-source model got delayed because Kimi K2 stole their thunder and beat their numbers.
57.
by fzysingularity 1y ago
This isn't surprising at all - most VLMs today are quite poor on localization even though they've been explicitly post-trained on object detection tasks. One insight that the author calls out is the inconsistencies in coordinate s
58.
by fzysingularity 1y ago
Nice!
59.
by fzysingularity 1y ago
The contributions for the Github project is quite intriguing: https://github.com/MiguelsPizza/WebMCP/graphs/contributors MiguelsPizza | 3 commits | 89++ | 410-- claude | 2 commits | 31,799++ | 0--
60.
by fzysingularity 1y ago
This demo showcases the latter approach with tool-calling - essentially filling in the gaps of current VLMs. That said, we're of course interested in folding all these capabilities into a single model, but that's going to take a b

More ›