# Ultralytics 🚀 AGPL-3.0 License - https://ultralytics.com/license

# Global configuration YAML with settings and hyperparameters for YOLO training, validation, prediction and export
# For documentation see https://docs.ultralytics.com/usage/cfg

task: detect # (str) YOLO task, i.e. detect, segment, semantic, depth, classify, pose, obb
mode: train # (str) YOLO mode, i.e. train, val, predict, export, track, benchmark

# Train settings -------------------------------------------------------------------------------------------------------
model: # (str, optional) path to model file, i.e. yolo26n.pt or yolo26n.yaml
data: # (str, optional) path to dataset file or classification directory/name
epochs: 100 # (int) number of epochs to train for
time: # (float, optional) max hours to train; overrides epochs if set
patience: 100 # (int) early stop after N epochs without val improvement
batch: 16 # (int | float) batch size as int (e.g. 16), -1 for AutoBatch, or float in (0.0, 1.0) for GPU memory fraction
imgsz: 640 # (int | list) train/val use int (square); predict/export may use [h,w]
save: True # (bool) save train checkpoints and predict results
save_period: -1 # (int) save checkpoint every N epochs; disabled if < 1
cache: False # (bool | str) cache images in RAM (True/'ram') or on 'disk' to speed dataloading; False disables
device: # (int | str | list) device: 0/[0,1] CUDA, 'npu:0' Ascend, 'xpu:0' Intel, 'cpu'/'mps', -1/[-1,-1] auto GPUs
workers: 8 # (int) dataloader workers (per RANK if DDP)
project: # (str, optional) project name for results root
name: # (str, optional) experiment name or export hardware target
exist_ok: False # (bool) overwrite existing 'project/name' if True
pretrained: True # (bool | str) use pretrained weights (bool) or load weights from path (str)
cls_remap: True # (bool) when fine-tuning across datasets, remap pretrained cls head rows by class-name match
optimizer: auto # (str) optimizer: SGD, MuSGD, Adam, Adamax, AdamW, NAdam, RAdam, RMSProp, or auto
verbose: True # (bool) print verbose logs
seed: 0 # (int) random seed for reproducibility
deterministic: True # (bool) enable deterministic ops; reproducible but may be slower
single_cls: False # (bool) treat all classes as a single class
rect: False # (bool) rectangular batches for train; rectangular batching for val when mode='val'
cos_lr: False # (bool) cosine learning rate scheduler
close_mosaic: 10 # (int) disable mosaic augmentation for final N epochs (0 to keep enabled)
resume: False # (bool) resume training from last checkpoint in the run dir
amp: True # (bool | str) train precision: True/'fp16' (AMP check), 'bf16', or False/'fp32'; AMP uses FP16 train validation
fraction: 1.0 # (float | int | list) train ratio/count or split values; 1 = all, integer >1 = count, test 0 = none
profile: False # (bool) profile ONNX/TensorRT speeds during training for loggers
freeze: # (int | list, optional) freeze first N layers (int), or specific layer indices or module names like "23.cv2" (list)
multi_scale: 0.0 # (float) multi-scale range as a fraction of imgsz; sizes are rounded to stride multiples
compile: False # (bool | str) enable torch.compile() backend='inductor'; True="default", False=off, or "default|reduce-overhead|max-autotune-no-cudagraphs"
channels_last: # (bool, optional) auto-enable for non-Windows CUDA training on PyTorch 1.11+ and supported x86 CPU inference

# Segmentation
overlap_mask: True # (bool) merge instance masks into one mask during training (segment only)
mask_ratio: 4 # (int) training mask downsample ratio (segment only), does not change predicted mask resolution

# Classification
dropout: 0.0 # (float) dropout for classification head (classify only)

# Val/Test settings ----------------------------------------------------------------------------------------------------
val: True # (bool) run validation/testing during training
split: val # (str) dataset split to evaluate: 'val', 'test' or 'train'
save_json: False # (bool) save COCO JSON (detect/segment/pose), compute detect size metrics, or save semantic PNG masks
conf: # (float, optional) confidence threshold; defaults: predict=0.25, val=0.001 (0.01 for obb)
iou: 0.7 # (float) IoU threshold used for NMS
max_det: 300 # (int) maximum number of detections per image
quantize: # (int|str, optional) precision: predict/val 16/fp16 or 32/fp32/None sets pt/torchscript compute; elsewhere the artifact/runtime decides, though 16 rounds openvino inputs; export also supports 8/int8 and w8a8|w16a16|w8a16|w8a32; train uses 8/int8 for quantization-aware training (QAT); replaces deprecated half/int8 args
dnn: False # (bool) use OpenCV DNN for ONNX inference
plots: True # (bool) save plots and images during train/val
nms: # (bool, optional) None: external NMS (default); True: embed NMS on export; False: NMS-free head if available

# Predict settings -----------------------------------------------------------------------------------------------------
source: # (str, optional) path/dir/URL/stream for images or videos; e.g. 'ultralytics/assets' or '0' for webcam
vid_stride: 1 # (int) read every Nth frame for video sources
stream_buffer: False # (bool) True buffers all frames; False keeps the most recent frame for low-latency streams
visualize: False # (bool) save class activation heatmaps (predict) or visualize TP/FP/FN confusion (val)
augment: False # (bool) apply test-time augmentation during prediction
agnostic_nms: False # (bool) class-agnostic NMS
classes: # (int | list[int], optional) filter by class id(s), e.g. 0 or [0,2,3]
retina_masks: False # (bool) use high-resolution segmentation masks (segment)
embed: # (list[int], optional) return feature embeddings from given layer indices

# Visualize settings ---------------------------------------------------------------------------------------------------
show: False # (bool) show images/videos in a window if supported
save_frames: False # (bool) save individual frames from video predictions
save_txt: False # (bool) save results as .txt files (xywh format)
save_conf: False # (bool) save confidence scores with results
save_crop: False # (bool) save cropped prediction regions to files
show_labels: True # (bool) draw class labels on images, e.g. 'person'
show_conf: True # (bool) draw confidence values on images, e.g. '0.99'
show_boxes: True # (bool) draw bounding boxes on images
line_width: # (int, optional) line width of boxes; auto-scales with image size if not set

# Export settings ------------------------------------------------------------------------------------------------------
format: torchscript # (str) target format, e.g. torchscript|onnx|openvino|engine|coreml|coreai|saved_model|pb|edgetpu|litert|paddle|mnn|ncnn|imx|rknn|executorch|axelera|deepx|qnn|hailo|ascend|xilinx
optimize: False # (bool) DEEPX only; higher compiler optimization (slower compile, faster inference)
dynamic: False # (bool) dynamic shapes for torchscript, onnx, openvino, engine, coreml, mnn; enable variable image sizes
simplify: True # (bool) supported ONNX-based exports (see export table); simplify intermediate ONNX graph
opset: # (int, optional) supported ONNX-based exports (see export table); intermediate ONNX opset version
workspace: # (float, optional) engine (TensorRT) only; workspace size in GiB, e.g. 4

# Hyperparameters ------------------------------------------------------------------------------------------------------
lr0: 0.01 # (float) initial learning rate (SGD=1e-2, Adam/AdamW=1e-3)
lrf: 0.01 # (float) final LR fraction; final LR = lr0 * lrf
momentum: 0.937 # (float) SGD momentum or Adam beta1
weight_decay: 0.0005 # (float) weight decay (L2 regularization)
warmup_epochs: 3.0 # (float) warmup epochs (fractions allowed)
warmup_momentum: 0.8 # (float) initial momentum during warmup
warmup_bias_lr: 0.1 # (float) bias learning rate during warmup
distill_model: # (str, optional) path to teacher model for knowledge distillation
dis: 6.0 # (float) distillation loss weight
box: 7.5 # (float) box loss gain
cls: 0.5 # (float) classification loss gain
cls_pw: 0.0 # (float) class weights power for handling class imbalance (0.0=disable, 1.0=full inverse frequency)
dfl: 1.5 # (float) box distance loss gain (DFL when reg_max > 1, L1 on DFL-free YOLO26)
pose: 12.0 # (float) pose loss gain (pose tasks)
kobj: 1.0 # (float) keypoint objectness loss gain (pose tasks)
rle: 1.0 # (float) rle loss gain (pose tasks)
angle: 1.0 # (float) oriented angle loss gain (obb tasks)
dlog: 1.0 # (float) depth log (SILog) loss gain (depth tasks)
dgrad: 0.5 # (float) depth gradient loss gain (depth tasks)
dlam: 1.0 # (float) depth SILog variance focus: 1.0=scale-invariant, 0.0=log-RMSE (depth tasks)
nbs: 64 # (int) nominal batch size used for loss normalization
hsv_h: 0.015 # (float) HSV hue augmentation fraction
hsv_s: 0.7 # (float) HSV saturation augmentation fraction
hsv_v: 0.4 # (float) HSV value (brightness) augmentation fraction
degrees: 0.0 # (float) rotation degrees (+/-)
translate: 0.1 # (float) translation fraction (+/-)
scale: 0.5 # (float | tuple) scale gain (+/-) as float, or explicit (min, max) tuple
shear: 0.0 # (float) shear degrees (+/-)
perspective: 0.0 # (float) perspective fraction (0–0.001 typical)
flipud: 0.0 # (float) vertical flip probability
fliplr: 0.5 # (float) horizontal flip probability
bgr: 0.0 # (float) RGB↔BGR channel swap probability
mosaic: 1.0 # (float) mosaic augmentation probability
mixup: 0.0 # (float) MixUp augmentation probability
cutmix: 0.0 # (float) CutMix augmentation probability
copy_paste: 0.0 # (float) segment/semantic/obb copy-paste fraction of eligible objects
copy_paste_mode: flip # (str) copy-paste strategy for segment/semantic/obb: flip or mixup
auto_augment: randaugment # (str) classification auto augmentation policy: randaugment, autoaugment, augmix
erasing: 0.4 # (float) random erasing probability for classification (0.0–1.0)

# Custom config.yaml ---------------------------------------------------------------------------------------------------
cfg: # (str, optional) path to a config.yaml that overrides defaults

# Tracker settings ------------------------------------------------------------------------------------------------------
tracker: tracktrack.yaml # (str) tracker config: botsort.yaml, bytetrack.yaml, ocsort.yaml, deepocsort.yaml, fasttrack.yaml, tracktrack.yaml
