when train my own dataset like AVA,the accurrcy is 0

04/14 16:14:40][INFO] ava_helper.py: 62: Finished loading image paths from: /home/lhc/trainava/frame_lists/train.csv
[04/14 16:14:40][INFO] ava_helper.py: 107: Finished loading annotations from: /home/lhc/trainava/ava_train_v2.2.csv, /home/lhc/trainava/ava_train_v2.2.csv, /home/lhc/trainava/ava_detection_train_boxes_and_labels_include_negative_v2.2.csv
[04/14 16:14:40][INFO] ava_helper.py: 110: Detection threshold: 0.8
[04/14 16:14:40][INFO] ava_helper.py: 111: Number of unique boxes: 23
[04/14 16:14:40][INFO] ava_helper.py: 112: Number of annotations: 7
[04/14 16:14:40][INFO] ava_helper.py: 158: 23 keyframes used.
[04/14 16:14:40][INFO] ava_dataset.py: 88: === AVA dataset summary ===
[04/14 16:14:40][INFO] ava_dataset.py: 89: Split: train
[04/14 16:14:40][INFO] ava_dataset.py: 90: Number of videos: 1
[04/14 16:14:40][INFO] ava_dataset.py: 94: Number of frames: 7
[04/14 16:14:40][INFO] ava_dataset.py: 95: Number of key frames: 23
[04/14 16:14:40][INFO] ava_dataset.py: 96: Number of boxes: 23.
[04/14 16:14:40][INFO] ava_helper.py: 62: Finished loading image paths from: /home/lhc/trainava/frame_lists/val.csv
[04/14 16:14:40][INFO] ava_helper.py: 107: Finished loading annotations from: /home/lhc/trainava/ava_detection_val_boxes_and_labels.csv
[04/14 16:14:40][INFO] ava_helper.py: 110: Detection threshold: 0.8
[04/14 16:14:40][INFO] ava_helper.py: 111: Number of unique boxes: 2
[04/14 16:14:40][INFO] ava_helper.py: 112: Number of annotations: 0
[04/14 16:14:40][INFO] ava_helper.py: 158: 2 keyframes used.
[04/14 16:14:40][INFO] ava_dataset.py: 88: === AVA dataset summary ===
[04/14 16:14:40][INFO] ava_dataset.py: 89: Split: val
[04/14 16:14:40][INFO] ava_dataset.py: 90: Number of videos: 1
[04/14 16:14:40][INFO] ava_dataset.py: 94: Number of frames: 7
[04/14 16:14:40][INFO] ava_dataset.py: 95: Number of key frames: 2
[04/14 16:14:40][INFO] ava_dataset.py: 96: Number of boxes: 2.
[04/14 16:14:40][INFO] ava_helper.py: 62: Finished loading image paths from: /home/lhc/trainava/frame_lists/train.csv
[04/14 16:14:40][INFO] ava_helper.py: 62: Finished loading image paths from: /home/lhc/trainava/frame_lists/val.csv
[04/14 16:14:40][INFO] train_net.py: 631: Start epoch: 1
[04/14 16:14:48][INFO] logging.py: 99: json_stats: {"_type": "train_iter", "cur_epoch": "1/20", "cur_iter": "10", "dt": 0.64406, "dt_data": 0.00039, "dt_net": 0.64367, "eta": "0:00:01", "loss": 0.71979, "lr": 0.01647, "mode": "train"}
[04/14 16:14:48][INFO] train_net.py: 692: Epoch 0 takes 7.93s. Epochs from 0 to 0 take 7.93s in average and 7.93s in median.
[04/14 16:14:48][INFO] train_net.py: 698: For epoch 0, each iteraction takes 0.72s in average. From epoch 0 to 0, each iteraction takes 0.72s in average.
[04/14 16:14:56][INFO] logging.py: 99: json_stats: {"_type": "train_iter", "cur_epoch": "2/20", "cur_iter": "10", "dt": 0.64921, "dt_data": 0.00029, "dt_net": 0.64892, "eta": "0:00:01", "loss": 0.69233, "lr": 0.03644, "mode": "train"}
[04/14 16:14:57][INFO] train_net.py: 692: Epoch 1 takes 7.82s. Epochs from 0 to 1 take 7.88s in average and 7.88s in median.
[04/14 16:14:57][INFO] train_net.py: 698: For epoch 1, each iteraction takes 0.71s in average. From epoch 0 to 1, each iteraction takes 0.72s in average.
[04/14 16:14:58][INFO] ava_eval_helper.py: 154: Evaluating with 2 unique GT frames.
[04/14 16:14:58][INFO] ava_eval_helper.py: 155: Evaluating with 2 unique detection frames
[04/14 16:14:58][INFO] ava_eval_helper.py: 297: AVA results wrote to detections_latest.csv
[04/14 16:14:58][INFO] ava_eval_helper.py: 298: took 0 seconds.
[04/14 16:14:58][INFO] ava_eval_helper.py: 297: AVA results wrote to groundtruth_latest.csv
[04/14 16:14:58][INFO] ava_eval_helper.py: 298: took 0 seconds.
{ 'PascalBoxes_PerformanceByCategory/AP@0.5IOU/up': 0.0,
'PascalBoxes_Precision/mAP@0.5IOU': 0.0}
[04/14 16:14:58][INFO] ava_eval_helper.py: 164: AVA eval done in 0.009727 seconds.
[04/14 16:14:58][INFO] logging.py: 99: json_stats: {"RAM": "10.20/15.55G", "_type": "val_epoch", "cur_epoch": "2", "gpu_mem": "3.07G", "map": 0.00000, "mode": "val"}
[04/14 16:15:06][INFO] logging.py: 99: json_stats: {"_type": "train_iter", "cur_epoch": "3/20", "cur_iter": "10", "dt": 0.65023, "dt_data": 0.00031, "dt_net": 0.64992, "eta": "0:00:01", "loss": 0.41744, "lr": 0.05642, "mode": "train"}
[04/14 16:15:06][INFO] train_net.py: 692: Epoch 2 takes 7.99s. Epochs from 0 to 2 take 7.91s in average and 7.93s in median.
[04/14 16:15:06][INFO] train_net.py: 698: For epoch 2, each iteraction takes 0.73s in average. From epoch 0 to 2, each iteraction takes 0.72s in average.
[04/14 16:15:14][INFO] logging.py: 99: json_stats: {"_type": "train_iter", "cur_epoch": "4/20", "cur_iter": "10", "dt": 0.64349, "dt_data": 0.00041, "dt_net": 0.64307, "eta": "0:00:01", "loss": 4.89473, "lr": 0.07639, "mode": "train"}
[04/14 16:15:15][INFO] train_net.py: 692: Epoch 3 takes 7.87s. Epochs from 0 to 3 take 7.90s in average and 7.90s in median.
[04/14 16:15:15][INFO] train_net.py: 698: For epoch 3, each iteraction takes 0.72s in average. From epoch 0 to 3, each iteraction takes 0.72s in average.
[04/14 16:15:16][INFO] ava_eval_helper.py: 154: Evaluating with 2 unique GT frames.
[04/14 16:15:16][INFO] ava_eval_helper.py: 155: Evaluating with 2 unique detection frames
[04/14 16:15:16][INFO] ava_eval_helper.py: 297: AVA results wrote to detections_latest.csv
[04/14 16:15:16][INFO] ava_eval_helper.py: 298: took 0 seconds.
[04/14 16:15:16][INFO] ava_eval_helper.py: 297: AVA results wrote to groundtruth_latest.csv
[04/14 16:15:16][INFO] ava_eval_helper.py: 298: took 0 seconds.
{ 'PascalBoxes_PerformanceByCategory/AP@0.5IOU/up': 0.0,
'PascalBoxes_Precision/mAP@0.5IOU': 0.0}
[04/14 16:15:16][INFO] ava_eval_helper.py: 164: AVA eval done in 0.002757 seconds.
[04/14 16:15:16][INFO] logging.py: 99: json_stats: {"RAM": "9.99/15.55G", "_type": "val_epoch", "cur_epoch": "4", "gpu_mem": "3.07G", "map": 0.00000, "mode": "val"}
[04/14 16:15:23][INFO] logging.py: 99: json_stats: {"_type": "train_iter", "cur_epoch": "5/20", "cur_iter": "10", "dt": 0.64331, "dt_data": 0.00032, "dt_net": 0.64299, "eta": "0:00:01", "loss": 30.00111, "lr": 0.09637, "mode": "train"}
[04/14 16:15:24][INFO] train_net.py: 692: Epoch 4 takes 7.78s. Epochs from 0 to 4 take 7.88s in average and 7.87s in median.
[04/14 16:15:24][INFO] train_net.py: 698: For epoch 4, each iteraction takes 0.71s in average. From epoch 0 to 4, each iteraction takes 0.72s in average.
[04/14 16:15:32][INFO] logging.py: 99: json_stats: {"_type": "train_iter", "cur_epoch": "6/20", "cur_iter": "10", "dt": 0.64864, "dt_data": 0.00044, "dt_net": 0.64820, "eta": "0:00:01", "loss": 50.00000, "lr": 0.10000, "mode": "train"}
[04/14 16:15:32][INFO] train_net.py: 692: Epoch 5 takes 8.13s. Epochs from 0 to 5 take 7.92s in average and 7.90s in median.
[04/14 16:15:32][INFO] train_net.py: 698: For epoch 5, each iteraction takes 0.74s in average. From epoch 0 to 5, each iteraction takes 0.72s in average.
[04/14 16:15:34][INFO] ava_eval_helper.py: 154: Evaluating with 2 unique GT frames.
[04/14 16:15:34][INFO] ava_eval_helper.py: 155: Evaluating with 2 unique detection frames
[04/14 16:15:34][INFO] ava_eval_helper.py: 297: AVA results wrote to detections_latest.csv
[04/14 16:15:34][INFO] ava_eval_helper.py: 298: took 0 seconds.
[04/14 16:15:34][INFO] ava_eval_helper.py: 297: AVA results wrote to groundtruth_latest.csv
[04/14 16:15:34][INFO] ava_eval_helper.py: 298: took 0 seconds.
{ 'PascalBoxes_PerformanceByCategory/AP@0.5IOU/up': 0.0,
'PascalBoxes_Precision/mAP@0.5IOU': 0.0}
[04/14 16:15:34][INFO] ava_eval_helper.py: 164: AVA eval done in 0.002891 seconds.
[04/14 16:15:34][INFO] logging.py: 99: json_stats: {"RAM": "9.91/15.55G", "_type": "val_epoch", "cur_epoch": "6", "gpu_mem": "3.07G", "map": 0.00000, "mode": "val"}
[04/14 16:15:41][INFO] logging.py: 99: json_stats: {"_type": "train_iter", "cur_epoch": "7/20", "cur_iter": "10", "dt": 0.64342, "dt_data": 0.00028, "dt_net": 0.64315, "eta": "0:00:01", "loss": 25.00000, "lr": 0.10000, "mode": "train"}
[04/14 16:15:42][INFO] train_net.py: 692: Epoch 6 takes 7.76s. Epochs from 0 to 6 take 7.90s in average and 7.87s in median.
[04/14 16:15:42][INFO] train_net.py: 698: For epoch 6, each iteraction takes 0.71s in average. From epoch 0 to 6, each iteraction takes 0.72s in average.
[04/14 16:15:49][INFO] logging.py: 99: json_stats: {"_type": "train_iter", "cur_epoch": "8/20", "cur_iter": "10", "dt": 0.65595, "dt_data": 0.00028, "dt_net": 0.65568, "eta": "0:00:01", "loss": 50.00000, "lr": 0.10000, "mode": "train"}