thi23459112

想請問一下 作者你的版本跟yolo11官方團隊的TensorRT 差異在哪 因為官方的也有內建 將權重pt轉成engine 因此才會想問倆著差異 還有 為何您的版本沒辦法建立int8 但官方的可以 謝謝

> [@Y-T-G](https://github.com/Y-T-G) Thank you for your reply. I changed the size from 640 to 2560. Since the dataset's annotations are imbalanced, is it possible to balance the dataset using Ultralytics'...

@Y-T-G I’d like to ask for some advice. Since the image is a panoramic photo, the resolution is very large. Some objects take up more than 50% of the frame,...

> You can use SAHI > > https://docs.ultralytics.com/guides/sahi-tiled-inference/ However, when using SAHI, it tends to mishandle some large objects, which increases misdetections.

@Y-T-G @glenn-jocher In my dataset, some objects occupy as much as 70–80% of the image. If I apply SAHI directly, those large objects are very likely to get cut across...

> Can you provide the example video? My situation is this: when processing a very long video—up to five hours in my tests—I consistently hit a “WARNING ⚠️ NMS time...

> Can you provide the example video? Hi, may I ask if there’s been any progress on this issue? I’ve noticed that the NMS LIMIT warning usually only shows up...

> Can you provide the example video? The issue has been identified. It is caused by loading the ReID weights from clip_vehicleid.pt, which triggers an NMS limit warning when processing...

> 1. Its not ultralytics's decision to make if yolov11 gets made, some researcher is probably already working on it > 2. The models are with a imgsz of 640...