yolov7_tiny_320_300e_coco.yml单类别目标检测，输出置信度在0.5左右

[JohnsenJiang]

问题确认 Search before asking

• 我已经搜索过问题，但是没有找到解答。I have searched the question and found no related answer.

请提出你的问题 Please ask your question

yolov7_tiny_320_300e_coco.yml单类别目标检测，将训练图像输入infer.py，输出置信度在0.5左右。

训练小部分log
[08/19 00:33:16] ppdet.engine INFO: Epoch: [997] [100/328] eta: 0:02:05 lr: 0.000007 loss_box: 0.012405 loss_cls: 0.000000 loss_obj: 0.016536 loss: 0.028852 batch_cost: 0.1540 data_cost: 0.0005 ips: 103.9102 images/s max_mem_reserved: 1701 MB d max_mem_allocated: 1228 MB
[08/19 00:33:31] ppdet.engine INFO: Epoch: [997] [200/328] eta: 0:01:50 lr: 0.000007 loss_box: 0.012056 loss_cls: 0.000000 loss_obj: 0.016032 loss: 0.027945 batch_cost: 0.1539 data_cost: 0.0004 ips: 103.9593 images/s max_mem_reserved: 1701 MB d max_mem_allocated: 1228 MB
[08/19 00:33:49] ppdet.engine INFO: Epoch: [997] [300/328] eta: 0:01:36 lr: 0.000007 loss_box: 0.012286 loss_cls: 0.000000 loss_obj: 0.016154 loss: 0.028774 batch_cost: 0.1775 data_cost: 0.0104 ips: 90.1479 images/s max_mem_reserved: 1701 MB d max_mem_allocated: 1228 MB

[JohnsenJiang]

coco_detection.yml中num_classes: 1
GPU为RTX4090D，paddle为develop版本，CUDA11.8，cudnn：8.6.0