Author here. D-FINE-seg is a real-time vision framework built on D-FINE (a DETR variant). One architecture, one config flag (task: detect | segment | sem_seg), five sizes N/S/M/L/X - switches between object detection, instance segmentation, and semantic segmentation.
The most recent part I built: the semantic seg head that outputs a dense mask, you can think of it as a per-pixel classification. No queries, no NMS, just a small conv neck + 1x1 classifier on full-frame features. It also exports to ONNX / TensorRT / OpenVINO / CoreML / LiteRT with a per-backend parity check.
On Cityscapes (TensorRT FP16, RTX 5070 Ti) it comes out ahead of YOLO26 and RF-DETR on detection/instance-segmentation F1 and takes the top mIoU on semantic segmentation, at real-time latency with 2–3x fewer params. I tried hard to make the benchmark difficult to game: every framework runs its own shipped inference code, each gets its own confidence threshold tuned for F1, and a single validator scores all of them against the same GT. Full protocol and every known asymmetry is public here: https://github.com/ArgoHA/cityscapes-benchmark.
It's not a fork - the detection core follows the D-FINE paper, but the segmentation heads, training, export and inference are implemented from scratch. You can take the code and train/finetune on your dataset.
Hi Peter. I am a Lead Computer Vision Engineer, working remotely at Veryfi (YC 17). I am a citizen of Russia, living in Armenia for last 3 years.
We applied to O1 visa, got approval from the USCIS, then got visa stamps and at the day of my flight visas were revoked. Embassy said to reapply - I did and now it's been 11 months of Administrative processing (as they said in the DoS).
My employer tried writing to the Congressman's office couple times, but every time Embassy answers ~"please wait, no timelines". I also filed a DHS Trip complaint and received this statement: "We have made any corrections to records that our inquiries determined were necessary, including, as appropriate, notations that may assist in avoiding incidents of misidentification".
With everything above, we were not able to get any meaningful timelines or figure what's going on. Is there anything we can do to make movement in my case? It's hard to plan your life living like this, so it is very important for me. Thanks!
If you want to try to take control back over this process (and your life), you probably need to file a mandamus action. This is essentially a demand that a government official execute his or her official duties. I don't handle mandamus actions but I'd be happy to refer you to some very good attorneys who do.
My advice would be to relax, enjoy Armenia, and assume you are not entering the U.S. in the next couple of years, for any reason. Administrative processing (I assume you actually mean 221(g) refusal) can easily take 1-2 years. The most extreme case I've heard of took 4 years.
In 2016-2020 writing to a congressmen could actually help: https://daniel.haxx.se/blog/2020/11/09/a-us-visa-in-937-days... , but not anymore. Last year I've heard about some success stories with U.S. courts. Still, it took for one O-1 person about two years from the initial visa interview.
I've also never heard of anyone getting any timeline estimations.
Just I quick note - writing to congressman may actually help and worth trying. My mom's greencard was stuck in 2023 due to some bs with documents in consulate with officer there non responding. After the letter by my senator I got a call from the officer over there and the issue resolved within a day. This is of course different from op's issue, but just wanted to note that it is worth trying in any case.
There are people who file collective mandamus lawsuits, so you might consider joining one of those groups.
However, from what I've seen, it could be a waste of money, as there isn't convincing evidence that it accelerates AP. These lawsuits usually take many months, and AP often resolves "by itself" before the lawsuit reaches a resolution.
It’s doubtful you’ll get any answers. Lawyers are of little help due to “consular nonreviewability” and your case clearly has hit some sort of national security flag. Since this was pre Trump II, it likely was a real human review not their new approach of automated revocations for poorly designed queries (a student was recently revoked due to fishing license citations).
As I understand, AP generally means Washington and not the consulate is actually the block/decision maker now, but they wouldn’t even officially tell you that. It could be with State or they could be waiting an opinion from someone. Picture some CIA agent who got stuck with a boring desk job and has a stack of thousands of visas to research.
The most recent part I built: the semantic seg head that outputs a dense mask, you can think of it as a per-pixel classification. No queries, no NMS, just a small conv neck + 1x1 classifier on full-frame features. It also exports to ONNX / TensorRT / OpenVINO / CoreML / LiteRT with a per-backend parity check.
On Cityscapes (TensorRT FP16, RTX 5070 Ti) it comes out ahead of YOLO26 and RF-DETR on detection/instance-segmentation F1 and takes the top mIoU on semantic segmentation, at real-time latency with 2–3x fewer params. I tried hard to make the benchmark difficult to game: every framework runs its own shipped inference code, each gets its own confidence threshold tuned for F1, and a single validator scores all of them against the same GT. Full protocol and every known asymmetry is public here: https://github.com/ArgoHA/cityscapes-benchmark.
It's not a fork - the detection core follows the D-FINE paper, but the segmentation heads, training, export and inference are implemented from scratch. You can take the code and train/finetune on your dataset.
Blog write-up on the semantic segmentation work: https://blog.agnify.ai/d-fine-seg-semantic-segmentation/. Happy to take criticism on the eval.
reply