channel element 3.2 is not allocated

## env:
- ubuntu 16.04
- GPU nvidia 2080ti
- cuda10.1
- pytorch 1.5.0
## begin training
```
[INFO: kinetics.py: 73]: Constructing Kinetics train...
[INFO: kinetics.py: 108]: Constructing kinetics dataloader (size: 1804) from /home/xxx/dataset/actionData/train_data/train.csv
[INFO: kinetics.py: 73]: Constructing Kinetics val...
[INFO: kinetics.py: 108]: Constructing kinetics dataloader (size: 399) from /home/xxx/dataset/actionData/train_data/val.csv
[INFO: train_net.py: 291]: Start epoch: 1
curve_epoch: 0
[INFO: logging.py: 66]: json_stats: {"_type": "train_iter", "epoch": "1/300", "eta": "2:35:53", "iter": "10/60", "loss": 1.091222, "lr": 0.009961, "mem": 3667, "time_diff": 0.519902, "top1_err": 61.666672, "top5_err": 31.666668}
channel element 3.2 is not allocated
```
when train data size is over 1000, there is message **channel element 3.2 is not allocated**, i do not know how to eliminate this problem.