1 Release published by 1 person

• v8.3.83 v8.3.83 - `ultralytics 8.3.83` Revert saturation and value augmentation to relative shift (#19515)

  published Mar 4, 2025

2 Pull requests merged by 1 person

• ultralytics 8.3.83 Revert saturation and value augmentation to relative shift

  #19515 merged Mar 4, 2025

• Fix batch description for validation

  #19504 merged Mar 4, 2025

6 Pull requests opened by 4 people

• Align code examples with Google-style

  #19496 opened Mar 3, 2025

• improved error handling in session.py

  #19522 opened Mar 5, 2025

• Fix reference docs

  #19528 opened Mar 5, 2025

• Disable `save_hybrid` for OBB and update docs

  #19531 opened Mar 5, 2025

• Add https://youtu.be/ILqMBah5ZvI to docs

  #19532 opened Mar 5, 2025

• Add extra step for installing `tkinter` for annotator tool

  #19534 opened Mar 5, 2025

10 Issues closed by 8 people

• save_hybrid in model.val() mode throws dimension error for OBB

  #19516 closed Mar 5, 2025

• OBB detection and keypoints

  #19512 closed Mar 5, 2025

• Support Exporting YOLOv8 to ONNX with IR Version 8

  #19498 closed Mar 4, 2025

• How to switch to transformers/clip from openai/clip

  #19472 closed Mar 4, 2025

• model.predict() onnxruntime

  #19497 closed Mar 4, 2025

• The metrics for the last epoch in TensorBoard are being overridden.

  #18001 closed Mar 4, 2025

• Count of images and backgrounds in yolov8

  #18021 closed Mar 4, 2025

• Different image shape used for inference in Pytorch weights and ONNX

  #19238 closed Mar 3, 2025

• Why is the result of multi-gpu training different from that of single-gpu training?

  #19356 closed Mar 3, 2025

• How can I deploy Yolov11-obb?

  #17900 closed Mar 3, 2025

27 Issues opened by 27 people

• Digits classification or detection? What is the best way?

  #19533 opened Mar 5, 2025

• I am training with the RT-DETR model. Why is the accuracy close to 0?

  #19530 opened Mar 5, 2025

• Inconsistent Results in Yolov8 prediction

  #19529 opened Mar 5, 2025

• About the txt saved

  #19527 opened Mar 5, 2025

• Now I have the results.csv file of YOLOv8 YOLOv9 YOLOv10 YOLO11 YOLOv12. May I ask how to display the PR curves of these five models in the same coordinate system?

  #19526 opened Mar 5, 2025

• input shape for training

  #19525 opened Mar 5, 2025

• Threading issue on Train and Val

  #19524 opened Mar 5, 2025

• I encountered a problem while using YOLOv11 for object detection. When I detect a relatively long object, it doesn't get detected fully.

  #19523 opened Mar 5, 2025

• How can I train by applying image padding?

  #19521 opened Mar 5, 2025

• Custom trained YOLOv8n model, converted into tflite doesn't detect any objects in Android Studio

  #19520 opened Mar 4, 2025

• model.train() returns None when training on multiple GPUs.

  #19519 opened Mar 4, 2025

• Metrics all 0 after TensorRT INT8 export for mode val, only INT8 ONNX performs well

  #19518 opened Mar 4, 2025

• TAL seems have some shortcomings.

  #19514 opened Mar 4, 2025

• pt model to an ONNX model, different result

  #19513 opened Mar 4, 2025

• How to create labels of segmentation object with inner profile?

  #19511 opened Mar 4, 2025

• inf frames of shapre 2304x1536 at 2.00fps

  #19510 opened Mar 4, 2025

• How to generate a onxx and a nms file ?

  #19509 opened Mar 4, 2025

• resume in YOLOv8

  #19508 opened Mar 4, 2025

• Does Yolov11 not support multi-disjoint-part segment dataset yet?

  #19507 opened Mar 4, 2025

• How to handle out of picture keypoints with YOLO pose

  #19506 opened Mar 3, 2025

• how could i build the environment via conda when using YOLOv8 OnnxRuntime C++

  #19505 opened Mar 3, 2025

• How to replace the YOLOv11backbone with SwinTransformer?

  #19503 opened Mar 3, 2025

• How to calculate coordinates from yolo format to yolov8-obb format?

  #19502 opened Mar 3, 2025

• Why is the Inference Speed of YOLOv8 ONNX Much Slower Compared to PyTorch (.pt) Model

  #19499 opened Mar 3, 2025

• Using Sahi with instance segmentation model

  #19494 opened Mar 2, 2025

• How to handle cascade yolo models?

  #19493 opened Mar 2, 2025

• Excluding Partially Visible Objects in YOLO11 on iOS

  #19492 opened Mar 2, 2025

33 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• The 8.3.46 have shows that 'package' not found

  #18086 commented on Mar 2, 2025 • 0 new comments

• Different behavior (NOT good) of yolov11n on Jetson Orin Nano Super

  #19134 commented on Mar 2, 2025 • 0 new comments

• I noticed that YOLOv5 can achieve a better model by adding background images. Can YOLOv11 also be operated in the same way

  #19460 commented on Mar 2, 2025 • 0 new comments

• Turn off early stopping for first "n" epochs

  #19490 commented on Mar 2, 2025 • 0 new comments

• Yolo12 batch size

  #19488 commented on Mar 2, 2025 • 0 new comments

• How can I allign my YOLOv8 detection results with my ground truth ROI image?

  #17991 commented on Mar 3, 2025 • 0 new comments

• yolo11 model.track, sometime no masks data

  #19459 commented on Mar 3, 2025 • 0 new comments

• YOLOv10 or YOLO11 Pruning, Masking and Fine-Tuning

  #19476 commented on Mar 3, 2025 • 0 new comments

• Questions about target tracking

  #19473 commented on Mar 3, 2025 • 0 new comments

• yolo11ndetection evaluation numbers

  #19369 commented on Mar 3, 2025 • 0 new comments

• Question about self.loss *= world_size in trainer

  #18616 commented on Mar 3, 2025 • 0 new comments

• Show classes results after each epochs

  #17862 commented on Mar 4, 2025 • 0 new comments

• The matrix multiplication in the post-processing stage of YOLOSEG is quite time-consuming when performed on the CPU of edge devices. Why not include this operation in the model during export and utilize the GPU for inference?

  #19471 commented on Mar 4, 2025 • 0 new comments

• Implementing Knowledge Distillation in YOLOv11: Training the Student Model with Teacher Predictions

  #19386 commented on Mar 4, 2025 • 0 new comments

• Training Time

  #19298 commented on Mar 5, 2025 • 0 new comments

• Bad result after convert to OpenVINO format and run on DLStreamer pipeline

  #19452 commented on Mar 5, 2025 • 0 new comments

• PytorchStreamReader failed reading zip archive: failed finding central directory / The amp parameter cannot be set to True

  #19469 commented on Mar 5, 2025 • 0 new comments

• Quantization Issue Yolov11/v10/v9 -- Inference time and memory usage deteriorated

  #19446 commented on Mar 5, 2025 • 0 new comments

• OKS sigma as hyperparameter

  #3526 commented on Mar 5, 2025 • 0 new comments

• Add explicit angle loss in OBB

  #13157 commented on Mar 5, 2025 • 0 new comments

• Support non-square size for classify. Use `imgsz` from metadata when available.

  #16630 commented on Mar 5, 2025 • 0 new comments

• faster-coco-eval support

  #17020 commented on Mar 5, 2025 • 0 new comments

• Configurable oks sigma

  #17274 commented on Mar 5, 2025 • 0 new comments

• Try uv pip install for RPi Benchmarks

  #17912 commented on Mar 3, 2025 • 0 new comments

• Add custom tensorboard logdir

  #18350 commented on Mar 5, 2025 • 0 new comments

• Improve solutions for better support 🚀

  #18491 commented on Mar 5, 2025 • 0 new comments

• Finer control over plotted prediction output by having more control variables Issue # 18418

  #18800 commented on Mar 3, 2025 • 0 new comments

• Fix result saving issue when multi-stream detection

  #19297 commented on Mar 3, 2025 • 0 new comments

• Update inference image path with assets in model docs

  #19403 commented on Mar 3, 2025 • 0 new comments

• Replaces use of Python builtins

  #19405 commented on Mar 2, 2025 • 0 new comments

• Add https://youtu.be/B5HvdWmF5g4 to docs

  #19479 commented on Mar 3, 2025 • 0 new comments

• Constrain OpenVINO versions to >=2024.0.0,!=2025.0.0

  #19481 commented on Mar 4, 2025 • 0 new comments

• Optimize RKNN export and inference structure

  #19482 commented on Mar 3, 2025 • 0 new comments