Skip to content

Improve training output #13

@edoyango

Description

@edoyango

The training progress updates are hard to parse e.g.:

     Epoch   gpu_mem       box       obj       cls     total    labels  img_size
                 all        1551      515206       0.051      0.0376     0.00252    0.000448
      1/99     27.3G   0.07209    0.4247         0    0.4968      2836       640: 100%|██████████| 291/291 [09:59<00:00,  2.06s/it]
               Class      Images      Labels           P           R      mAP@.5  mAP@.5:.95: 100%|██████████| 49/49 [00:35<00:00,  1.37it/s]

Columns are unaligned, and rows seem interleaved? Ideally it would look more like the latest yolo output:

      Epoch    GPU_mem   box_loss   cls_loss   dfl_loss  Instances       Size
    300/300      22.3G      1.347     0.5884      1.779          0        864: 100%|██████████| 94/94 [00:45<00:00,  2.06it/s]
                 Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100%|██████████| 30/30 [00:06<00:00,  4.59it/s]
                   all        943       1497      0.934      0.925      0.951      0.645

tbf, the yolo output could also be improved a little.

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions