shooorf

@eloquentarduino latest tensorflow lite micro has 97 builtin operators in AllOpsResolver Do you have any plans to upgrade EloquentTinyML to this new version? Thanks for a great job you've done...

Same bug. Model trained and loaded from checkpoint with ``` trainer = Trainer(experiment_name='my_first_yolonas_run', ckpt_root_dir='checkpoints') train_data = coco_detection_yolo_format_train(dataset_params=..., dataloader_params=...) val_data = coco_detection_yolo_format_val(dataset_params=..., dataloader_params=...) trainer.train(model=model, training_params=train_params, train_loader=train_data, valid_loader=val_data) ... models.get(Models.YOLO_NAS_S,num_classes=len(dataset_params['classes']),checkpoint_path='checkpoints/my_first_yolonas_run/ckpt_best.pth') ``` has...

**Stupid Workaround:** File /usr/local/lib/python3.8/dist-packages/super_gradients/training/pipelines/pipelines.py (change path to your environment specific) Function `_generate_prediction_result_single_batch` Line 175: `preprocessed_images.append(preprocessed_image)` Change to `preprocessed_images.append(preprocessed_image.astype(np.single))` If you used `coco_detection_yolo_format_train` and `coco_detection_yolo_format_val `functions from example notebook: you model...

> @shooorf how can I see that list of preprocessings from my checkpoint ? from super_gradients.training.utils.checkpoint_utils import read_ckpt_state_dict stdict = read_ckpt_state_dict(ckpt_path=checkpoint_path) print(stdict['processing_params']['image_processor'])