Validation metrics = 0.0 during training yolo-nas

[Zat-Code]

💡 Your Question

Hi !

i want fine-tuning yolo-nas-m.

i use coco_detection_yolo_format_train and coco_detection_yolo_format_val for load my dataset (in Darknet/Yolo format).

when i do :

self.val_data.dataset.get_random_sample(), the console display me a coherent data in coco format.

i have already use this dataset for train yolov7 model and it's work.

training_params = {
            'silent_mode': False,
            'average_best_models': True,
            'warmup_mode': "LinearEpochLRWarmup",
            'warmup_initial_lr': 0.00001,
            'lr_warmup_epochs': 3,
            'initial_lr': {"backbone": 0.01, "default": 0.1},
            'lr_mode': "cosine",
            'cosine_final_lr_ratio': 0.1,
            'optimizer': "Adam",
            'optimizer_params': {weight_decay: 0.0001},
            'zero_weight_decay_on_bias_and_bn': True,
            'ema': True,
            'ema_params:' {"decay": 0.9, "decay_type": "threshold"},
            "mixed_precision": True,
            "max_epochs": self.epochs,
            "transforms": self.load_transform_param(),
            "loss": PPYoloELoss(
                use_static_assigner=False,
                num_classes=len(self.classes),
                reg_max=16
            ),
            "train_metrics_list": [ 
            ],
            "valid_metrics_list": [
                DetectionMetrics_050(
                    score_thres=0.1,
                    top_k_predictions=300,
                    num_cls=len(self.classes),
                    normalize_targets=True,
                    calc_best_score_thresholds=True,
                    post_prediction_callback=PPYoloEPostPredictionCallback(
                        score_threshold=0.01,
                        nms_top_k=256,
                        max_predictions=100,
                        nms_threshold=0.7
                        )
            )],
            "metric_to_watch": 'F1@0.50',
            "greater_metric_to_watch_is_better": True,
            "phase_callbacks" : 
                [
                ExtremeBatchDetectionVisualizationCallback(
                    metric=DetectionMetrics_050(
                    score_thres=0.1,
                    top_k_predictions=300,
                    num_cls=len(self.classes),
                    normalize_targets=True,
                    post_prediction_callback=PPYoloEPostPredictionCallback(
                        score_threshold=0.01,
                        nms_top_k=1000,
                        max_predictions=300,
                        nms_threshold=0.7
                        )
                    ),
                    classes=self.classes,
                    metric_component_name = 'mAP@0.50',
                    post_prediction_callback=PPYoloEPostPredictionCallback(
                        score_threshold=0.25,
                        nms_top_k=1000,
                        max_predictions=300,
                        nms_threshold=0.7
                        ),
                    normalize_targets=True
                )
            ]
        }

when i launch my training, this is the prompt cmd :

after a bit of searching, I saw that this could be due to a dataset that has no positive objects.
Of the two classes in my dataset, I've only annotated my second class. Could this be the cause of my problem?

on yolov7 I have no problem, and I've indicated in my dataset_params the 'class_inclusion_list' argument with my second class

Thank you in advance for an answer !

Versions

No response