why mAP of my train model is under mAP of this paper？

[MillX2021]

I download Resnet101 GFocal model from mmdetction（download link：https://download.openmmlab.com/mmdetection/v2.0/gfl/gfl_r101_fpn_mstrain_2x_coco/gfl_r101_fpn_mstrain_2x_coco_20200629_200126-dd12f847.pth), and train by this command line： bash ./tools/dist_train.sh configs/ld/ld_gflv1_r101_r50_fpn_coco_1x.py 8 but I get the mAP of epoch 12 is 0.3960。this paper give this mAP should be 41.2。 train and eval log： 2021-08-18 10:02:25,727 - mmdet - INFO - Epoch [12][4750/4887] lr: 3.750e-05, eta: 0:02:01, time: 0.882, data_time: 0.015, memory: 6289, loss_cls: 0.3524, loss_bbox: 0.2883, loss_ld: 0.1610, loss_dfl: 0.2044, loss: 1.0061 2021-08-18 10:03:10,220 - mmdet - INFO - Epoch [12][4800/4887] lr: 3.750e-05, eta: 0:01:17, time: 0.890, data_time: 0.014, memory: 6289, loss_cls: 0.3597, loss_bbox: 0.2918, loss_ld: 0.1602, loss_dfl: 0.2045, loss: 1.0162 2021-08-18 10:03:54,650 - mmdet - INFO - Epoch [12][4850/4887] lr: 3.750e-05, eta: 0:00:32, time: 0.888, data_time: 0.015, memory: 6289, loss_cls: 0.3543, loss_bbox: 0.2908, loss_ld: 0.1611, loss_dfl: 0.2060, loss: 1.0122 2021-08-18 10:04:45,038 - mmdet - INFO - Saving checkpoint at 12 epochs [>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>] 5000/5000, 95.6 task/s, elapsed: 52s, ETA: 0s

2021-08-18 10:05:50,605 - mmdet - INFO - Evaluating bbox... Loading and preparing results... DONE (t=6.00s) creating index... index created! Running per image evaluation... Evaluate annotation type bbox DONE (t=53.03s). Accumulating evaluation results... DONE (t=11.64s). Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.396 Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=1000 ] = 0.572 Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=1000 ] = 0.428 Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=1000 ] = 0.226 Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=1000 ] = 0.435 Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=1000 ] = 0.516 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.582 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=300 ] = 0.582 Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=1000 ] = 0.582 Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=1000 ] = 0.369 Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=1000 ] = 0.630 Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=1000 ] = 0.745 2021-08-18 10:07:02,558 - mmdet - INFO - Exp name: ld_gflv1_r101_r50_fpn_coco_1x.py 2021-08-18 10:07:02,559 - mmdet - INFO - Epoch(val) [12][4887] bbox_mAP: 0.3960, bbox_mAP_50: 0.5720, bbox_mAP_75: 0.4280, bbox_mAP_s: 0.2260, bbox_mAP_m: 0.4350, bbox_mAP_l: 0.5160, bbox_mAP_copypaste: 0.396 0.572 0.428 0.226 0.435 0.516

could you help me to slove this problem？

[HikariTJU]

You should edit config file according your gpu num and samples_per_gpu, please refer to "Learning rate setting" part in readme.

For 8 gpus, I recommend using samples_per_gpu=2 lr=0.01

[MillX2021]

thanks. I'm going to recurrence it.

[Zzh-tju]

I download Resnet101 GFocal model from mmdetction（download link：https://download.openmmlab.com/mmdetection/v2.0/gfl/gfl_r50_fpn_mstrain_2x_coco/gfl_r50_fpn_mstrain_2x_coco_20200629_213802-37bb1edc.pth)

Are you sure you download R101 GFocal model？

[MillX2021]

I download Resnet101 GFocal model from mmdetction（download link：https://download.openmmlab.com/mmdetection/v2.0/gfl/gfl_r50_fpn_mstrain_2x_coco/gfl_r50_fpn_mstrain_2x_coco_20200629_213802-37bb1edc.pth)

Are you sure you download R101 GFocal model？

hhh, i quote error.but I download r1010 gfocal model from https://download.openmmlab.com/mmdetection/v2.0/gfl/gfl_r101_fpn_mstrain_2x_coco/gfl_r101_fpn_mstrain_2x_coco_20200629_200126-dd12f847.pth. but I get the mAP is 40.8(the mAP is 41.2 in this paper).

[Zzh-tju]

@MillX2021 41.2 AP in the paper is evaluated on coco test-dev 2019. For val2017, AP is 41.1