why trained model can't detect objects correctly?

[StrugglingForBetter]

@chuanqi305 Thanks for your contribution.

We used mobile net v2(generated all prototxt files following your steps) to train our own dataset. After more than 80000 iterations, its MAP exceeds 70%. But when we use deploy.prototxt and caffemodel to do test, it can't detect objects correctly. It will show a lot of blocks which has very high confidence like 99.5% even 100%. Those blocks are mostly small rectangular squares.

Is there anyboy who came across the same issue？ How to solve this issue?

Thanks!

Below is my solver for training:

train_net: "train.prototxt" test_net: "test.prototxt" test_iter: 673 base_lr: 0.0005 display: 10 test_interval: 2000 max_iter: 120000 lr_policy: "multistep" gamma: 0.5 weight_decay: 0.00001 snapshot: 1000 snapshot_prefix: "snapshot/mobilenet" solver_mode: GPU debug_info: false snapshot_after_train: true test_initialization: false average_loss: 10 stepvalue: 20000 stepvalue: 40000 stepvalue: 80000 stepvalue: 110000 iter_size: 1 type: "RMSProp" eval_type: "detection" ap_version: "11point"