关于半监督中指标不正确的情况

[lzd-1230]

在半监督训练开启之后出现了指标掉0，并且val_loss这边一直为0

               epoch,      train/box_loss,      train/obj_loss,      train/cls_loss,   metrics/precision,      metrics/recall,     metrics/mAP_0.5,metrics/mAP_0.5:0.95,        val/box_loss,        val/obj_loss,        val/cls_loss,               x/lr0,               x/lr1,               x/lr2
                   0,            0.026734,           0.0030149,           0.0025407,             0.51456,             0.33336,             0.31076,             0.14092,                   0,                   0,                   0,               6e-05,               6e-05,             0.09946
                   1,            0.024462,           0.0026633,           0.0020524,             0.51016,             0.32951,             0.30117,             0.13794,                   0,                   0,                   0,          0.00012969,          0.00012969,             0.09883
                   2,            0.026195,           0.0026313,           0.0023474,             0.42978,             0.33882,             0.29534,             0.13781,                   0,                   0,                   0,          0.00019905,          0.00019905,            0.098199
                   3,            0.027825,           0.0031602,           0.0024128,             0.51994,             0.30863,             0.29933,             0.13462,                   0,                   0,                   0,          0.00026808,          0.00026808,            0.097568
                   4,            0.024026,           0.0030408,           0.0023448,              0.5138,             0.32009,              0.2979,             0.13442,                   0,                   0,                   0,          0.00033677,          0.00033677,            0.096937
                   5,            0.025969,           0.0035441,           0.0021407,             0.49064,             0.32992,             0.30062,              0.1365,                   0,                   0,                   0,          0.00040513,          0.00040513,            0.096305
                   6,            0.024222,           0.0030058,           0.0020596,             0.42509,             0.34826,             0.29972,             0.13865,                   0,                   0,                   0,          0.00047316,          0.00047316,            0.095673
                   7,            0.024484,           0.0025748,           0.0023783,             0.49233,             0.32885,             0.30761,             0.13922,                   0,                   0,                   0,          0.00054086,          0.00054086,            0.095041
                   8,            0.025334,            0.002859,           0.0020104,             0.50474,             0.32826,             0.30648,              0.1426,                   0,                   0,                   0,          0.00060822,          0.00060822,            0.094408
                   9,            0.025303,           0.0029177,            0.002478,             0.49028,             0.32404,             0.30023,             0.13311,                   0,                   0,                   0,          0.00067525,          0.00067525,            0.093775
                  10,            0.038496,           0.0039158,           0.0042909,             0.49051,             0.32394,             0.29715,             0.13195,                   0,                   0,                   0,           0.0043931,           0.0043931,            0.059393
                  11,            0.044599,           0.0045264,           0.0069123,             0.51159,             0.31886,             0.30169,               0.132,                   0,                   0,                   0,           0.0047817,           0.0047817,            0.055682
                  12,            0.046952,           0.0051121,           0.0058509,             0.47838,             0.33643,             0.30369,             0.13104,                   0,                   0,                   0,           0.0051684,           0.0051684,            0.051968
                  13,            0.062479,           0.0087104,            0.012121,             0.53358,             0.32949,              0.3299,             0.13967,                   0,                   0,                   0,           0.0055531,           0.0055531,            0.048253
                  14,            0.094865,            0.020805,            0.024056,                   0,                   0,                   0,                   0,                   0,                   0,                   0,           0.0059359,           0.0059359,            0.044536
                  15,            0.097925,            0.012116,            0.024947,                   0,                   0,                   0,                   0,                   0,                   0,                   0,           0.0063167,           0.0063167,            0.040817
                  16,            0.090215,           0.0061514,            0.024214,                   0,                   0,                   0,                   0,                   0,                   0,                   0,           0.0066956,           0.0066956,            0.037096
                  17,            0.088175,           0.0061138,            0.022982,                   0,                   0,                   0,                   0,                   0,                   0,                   0,           0.0070725,           0.0070725,            0.033372
                  18,            0.087877,           0.0059195,            0.022877,                   0,                   0,                   0,                   0,                   0,                   0,                   0,           0.0074475,           0.0074475,            0.029647
                  19,            0.085309,           0.0059199,            0.023373,                   0,                   0,                   0,                   0,                   0,                   0,                   0,           0.0078205,           0.0078205,             0.02592
                  20,            0.086314,           0.0063424,            0.022941,                   0,                   0,                   0,                   0,                   0,                   0,                   0,           0.0081916,           0.0081916,            0.022192
                  21,             0.08436,           0.0058643,            0.022787,                   0,                   0,                   0,                   0,                   0,                   0,                   0,           0.0085607,           0.0085607,            0.018461

麻烦大神帮忙看下是否是配置有问题？

# EfficientTeacher by Alibaba Cloud 

# Parameters
project: './runs_yolov5'
adam: False
epochs: 380
weights: '/efficientteacher/runs_yolov5/exp3/weights/best.pt'
prune_finetune: False
linear_lr: True
hyp:
  lr0: 0.01
  hsv_h: 0.015
  hsv_s: 0.7
  hsv_v: 0.4
  lrf: 0.1
  scale: 0.9
  no_aug_epochs: 0
  mixup: 0.1
  warmup_epochs: 3
  burn_epochs: 10  # 添加burn_epochs参数

Model:
  depth_multiple: 1.00  # model depth multiple
  width_multiple: 1.00  # layer channel multiple
  Backbone: 
    name: 'YoloV5'
    activation: 'SiLU'
  Neck: 
    name: 'YoloV5'
    in_channels: [256, 512, 1024]
    out_channels: [256, 512, 1024]
    activation: 'SiLU'
  Head: 
    name: 'YoloV5'
    activation: 'SiLU'
  anchors: [[10,13, 16,30, 33,23],[30,61, 62,45, 59,119],[116,90, 156,198, 373,326]]  # P5/32]
Loss:
  type: 'ComputeLoss'
  cls: 0.3
  obj: 0.7
  anchor_t: 4.0

Dataset:
  data_name: 'gray_sample_raw'
  train: /home/data/lzd/honor_dataset/gray_sample_raw/train/train.txt
  val: /home/data/lzd/honor_dataset/gray_sample_raw/valid/val.txt
  test: data/custom_val.txt # 20288 of 40670 images, submit to https://competitions.codalab.org/competitions/20794^
  target: /home/data/lzd/honor_dataset/vis_good/train/unlabel.txt
  nc: 4  # number of classes
  np: 0 #number of keypoints
  names: ['aa', 'bb', 'cc', 'dd']
  img_size: 768
  # 总共的batch_size
  batch_size: 128

SSOD:
  train_domain: True
  nms_conf_thres: 0.4
  nms_iou_thres: 0.5
  teacher_loss_weight: 1.0
  cls_loss_weight: 0.3
  box_loss_weight: 0.05
  obj_loss_weight: 0.7
  loss_type: 'ComputeStudentMatchLoss'
  ignore_thres_low: 0.2
  ignore_thres_high: 0.4
  uncertain_aug: True
  use_ota: False
  multi_label: False
  ignore_obj: False
  pseudo_label_with_obj: True
  pseudo_label_with_bbox: True
  pseudo_label_with_cls: False
  with_da_loss: False
  da_loss_weights: 0.01
  epoch_adaptor: True
  resample_high_percent: 0.25
  resample_low_percent: 0.99
  ema_rate: 0.999
  cosine_ema: True
  imitate_teacher: False
  ssod_hyp:
    with_gt: False
    mosaic: 1.0
    cutout: 0.5
    autoaugment: 0.5
    scale: 0.8
    degrees: 0.0
    shear: 0.0

[kv1830]

（1）
epochs: 380
burn_epochs: 10 ----这个10相对380，可能太少了，建议设成epochs的一半或三分之一

（2）
ignore_thres_low: 0.2
ignore_thres_high: 0.4

可以用原来yolov5的配置试试，0.4有点低
ignore_thres_low: 0.1
ignore_thres_high: 0.6

（3）
确实你改了哪些代码，比如EMA其实是这里面相当重要的一环，我当时把EMA rate给调的很低（好像是0.3左右），开启伪标签后，两轮就飞。。。