This is not an issues< just a question>

[VYRION-Ai]

i do training using

!python train.py --batch-size 45 --img 448 448 --data '../data.yaml' --cfg cfg/yolor_p6.cfg --device 0 --hyp '/content/yolor/data/hyp.scratch.1280.yaml' --epochs 50

and here is the first Four Epochs

     Epoch   gpu_mem       box       obj       cls     total   targets  img_size
      0/49     8.55G    0.0194   0.03467  0.003309   0.05738        12       448: 100% 1626/1626 [22:57<00:00,  1.18it/s]

     Epoch   gpu_mem       box       obj       cls     total   targets  img_size
      1/49     11.9G   0.01943   0.03224   0.00101   0.05269        13       448: 100% 1626/1626 [22:55<00:00,  1.18it/s]

     Epoch   gpu_mem       box       obj       cls     total   targets  img_size
      2/49     11.9G   0.02142   0.03452  0.001187   0.05713        23       448: 100% 1626/1626 [22:56<00:00,  1.18it/s]

     Epoch   gpu_mem       box       obj       cls     total   targets  img_size
      3/49     11.9G   0.02173     0.035   0.00129   0.05802        13       448: 100% 1626/1626 [22:53<00:00,  1.18it/s]
               Class      Images        Targets            P               R         mAP@.5   mAP@.5:.95: 100% 350/350 [04:06<00:00,  1.42it/s]
                   all      3.14e+04    3.28e+04       0.603       0.861       0.829       0.574

my question is Are these result normal, the first three Epochs there is no learning , and then mAP@.5 --> 0.829? and how many Epochs i should train i am now in Epoch 10 and this are the results

 Epoch   gpu_mem       box       obj       cls     total   targets  img_size
      5/49     11.9G   0.01948   0.03259 0.0009156   0.05298        17       448: 100% 1626/1626 [22:53<00:00,  1.18it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 350/350 [04:01<00:00,  1.45it/s]
                 all    3.14e+04    3.28e+04       0.651       0.868       0.837       0.598

     Epoch   gpu_mem       box       obj       cls     total   targets  img_size
      6/49     11.9G   0.01893   0.03188 0.0008605   0.05167        15       448: 100% 1626/1626 [22:55<00:00,  1.18it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 350/350 [04:01<00:00,  1.45it/s]
                 all    3.14e+04    3.28e+04       0.623       0.872       0.841       0.601

     Epoch   gpu_mem       box       obj       cls     total   targets  img_size
      7/49     11.9G   0.01864   0.03138 0.0008095   0.05083        21       448: 100% 1626/1626 [22:54<00:00,  1.18it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 350/350 [04:02<00:00,  1.44it/s]
                 all    3.14e+04    3.28e+04       0.664       0.888       0.855       0.615

     Epoch   gpu_mem       box       obj       cls     total   targets  img_size
      8/49     11.9G   0.01835   0.03132 0.0007825   0.05046        16       448: 100% 1626/1626 [22:54<00:00,  1.18it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 350/350 [04:01<00:00,  1.45it/s]
                 all    3.14e+04    3.28e+04       0.671        0.89       0.856       0.615

     Epoch   gpu_mem       box       obj       cls     total   targets  img_size
      9/49     11.9G   0.01803   0.03087 0.0007548   0.04966        17       448: 100% 1626/1626 [22:54<00:00,  1.18it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 350/350 [04:01<00:00,  1.45it/s]
                 all    3.14e+04    3.28e+04       0.662       0.888       0.851       0.612

     Epoch   gpu_mem       box       obj       cls     total   targets  img_size
     10/49     11.9G   0.01788   0.03063 0.0007424   0.04925        13       448: 100% 1626/1626 [22:55<00:00,  1.18it/s]
               Class      Images     Targets           P           R      mAP@.5  mAP@.5:.95: 100% 350/350 [04:01<00:00,  1.45it/s]
                 all    3.14e+04    3.28e+04        0.66       0.889       0.851       0.612

[WongKinYiu]

it seems normal, if you want to test at the first 3 epochs, modify https://github.com/WongKinYiu/yolor/blob/main/train.py#L335

[VYRION-Ai]

But if i want to get 95% , how epoch should I train

[VYRION-Ai]

Or we could get to, let's say 96% or something higher. Is more training a benefit 50 epoch , 100 epoch or what?