training_config { # checkpoint: "/workspace/tao-experiments/efficientdet/pretrained_efficientdet_vefficientnet_b0/efficientnet_b0.hdf5" train_batch_size: 4 iterations_per_loop: 10 checkpoint_period: 10 #this is number of training images/gpus 18800/3 #18800 is the number of training images that have d/c labels. This was the output after running the convert_kitti_labels_to_coco cell num_examples_per_epoch: 6266 #num_examples_per_epoch: 6645 num_epochs: 300 tf_random_seed: 42 lr_warmup_epoch: 5 lr_warmup_init: 0.00005 learning_rate: 0.005 amp: True moving_average_decay: 0.9999 l2_weight_decay: 0.00004 l1_weight_decay: 0.0 } dataset_config { #num_classes is n+1 because there is a background class num_classes: 3 image_size: "1376,1024" training_file_pattern: "/workspace/taov3/trainingConfigs/ObjectDetection/efficientDet/efficientdet_initial_training/test_tf_records/train-*" validation_file_pattern: "/workspace/taov3/trainingConfigs/ObjectDetection/efficientDet/efficientdet_initial_training/test_tf_records/val-*" validation_json_file: "/workspace/taov3/trainingConfigs/ObjectDetection/efficientDet/efficientdet_initial_training/specs/val_coco.json" max_instances_per_image: 100 skip_crowd_during_training: True } model_config { model_name: 'efficientdet-d0' min_level: 3 max_level: 7 num_scales: 3 } augmentation_config { rand_hflip: False random_crop_min_scale: 1 random_crop_min_scale: 1 } eval_config { eval_batch_size: 4 eval_epoch_cycle: 2 #the number of validation images that have d/c labels. This was the output after running the convert_kitti_labels_to_coco cell eval_samples: 587 #eval_samples: 624 min_score_thresh: 0.4 max_detections_per_image: 100 }