TridentNet调试笔记

运行 create_voc_roidb.py

报错 No such file or directory: 'data/label_map/voc_label_map.json'

解决方法：https://github.com/TuSimple/simpledet/issues/282

 

运行：python detection_train.py --config config/faster_r50v1_fpn_1x.py

报错:Check failed: e == cudaSuccess || e == cudaErrorCudartUnloading CUDA: invalid device ordinal #15

解决方法：https://github.com/TuSimple/simpledet/issues/15

修改gpus和 loader_worker

class KvstoreParam:
        kvstore     = "local"
        batch_image = General.batch_image
        gpus        = [0, 1, 2, 3, 4, 5, 6, 7]
        fp16        = General.fp16

训练自己的数据集

运行

python detection_train.py --config config/faster_r50v1_fpn_1x.py

报错：断言失败，显示image_url不存在

assert os.path.exists(image_url)

解决办法：

filename = root.find("filename").text  改为  filename = root.find("filename").text+'.jpg'

成功运行

10-27 19:13:49 Epoch[5] Batch [690]     Iter: 4170/4176 Lr: 0.00250     Speed: 2.36 samples/sec Train-RpnAcc=0.999308,  RpnL1=0.153242, RcnnAcc=0.990205,     RcnnL1=0.548991,        
10-27 19:13:52 Epoch[5] Train-RpnAcc=0.999307
10-27 19:13:52 Epoch[5] Train-RpnL1=0.152955
10-27 19:13:52 Epoch[5] Train-RcnnAcc=0.990232
10-27 19:13:52 Epoch[5] Train-RcnnL1=0.548295
10-27 19:13:52 Epoch[5] Time cost=462.154
10-27 19:13:53 Saved checkpoint to "experiments/faster_r50v1_fpn_1x/checkpoint-0006.params"
10-27 19:13:53 Training has done
10-27 19:14:03 Exiting

其中

10-27 19:13:52 Epoch[5] Train-RpnAcc=0.999307  
10-27 19:13:52 Epoch[5] Train-RpnL1=0.152955
10-27 19:13:52 Epoch[5] Train-RcnnAcc=0.990232
10-27 19:13:52 Epoch[5] Train-RcnnL1=0.548295