super-gradients/src/super_gradients/recipes/training_hyperparams/coco2017_yolox_train_params.yaml

defaults:
  - default_train_params

max_epochs: 300
lr_mode: cosine
cosine_final_lr_ratio: 0.05
lr_warmup_epochs: 5
lr_cooldown_epochs: 15
initial_lr:  0.02
zero_weight_decay_on_bias_and_bn: True
batch_accumulate: 1

save_ckpt_epoch_list: [285]

loss: yolox_loss

criterion_params:
  strides: [8, 16, 32]  # output strides of all yolo outputs
  num_classes: 80

loss_logging_items_names: [iou, obj, cls, l1, num_fg, Loss]

optimizer: SGD
optimizer_params:
  momentum: 0.9
  weight_decay: 0.0005
  nesterov: True

ema: True
mixed_precision: True

valid_metrics_list:
  - DetectionMetrics:
      normalize_targets: True
      post_prediction_callback:
        _target_: super_gradients.training.models.detection_models.yolo_base.YoloPostPredictionCallback
        iou: 0.65
        conf: 0.01
      num_cls: 80

pre_prediction_callback: DetectionMultiscalePrePredictionCallback

phase_callbacks:
  - YoloXTrainingStageSwitchCallback:
      next_stage_start_epoch: 285

metric_to_watch: 'mAP@0.50:0.95'
greater_metric_to_watch_is_better: True


_convert_: all