End-to-end style train accuracy improves too quickly, am I wrong somewhere?

[Suncheng2022]

This is the first few iterations training log：

2023-04-23 16:36:09,169 - mmdet - INFO - workflow: [('train', 1)], max: 12 epochs
2023-04-23 16:36:30,358 - mmcv - INFO - Reducer buckets have been rebuilt in this iteration.
2023-04-23 16:37:03,128 - mmdet - INFO - Epoch [1][50/4399] lr: 1.978e-03, eta: 15:48:28, time: 1.079, data_time: 0.408, memory: 7504, loss_rpn_cls: 0.6129, loss_rpn_bbox: 0.3645, loss_cls: 113.8634, acc: 51.2393, loss_bbox: 0.4169, loss: 115.2578, grad_norm: 1112.3063
2023-04-23 16:37:36,244 - mmdet - INFO - Epoch [1][100/4399] lr: 3.976e-03, eta: 12:44:35, time: 0.662, data_time: 0.006, memory: 7504, loss_rpn_cls: 0.3718, loss_rpn_bbox: 0.2817, loss_cls: 0.5315, acc: 86.3486, loss_bbox: 0.4726, loss: 1.6576, grad_norm: 10.5886
2023-04-23 16:38:09,524 - mmdet - INFO - Epoch [1][150/4399] lr: 5.974e-03, eta: 11:43:52, time: 0.666, data_time: 0.006, memory: 7504, loss_rpn_cls: 0.2926, loss_rpn_bbox: 0.2397, loss_cls: 0.4812, acc: 85.5732, loss_bbox: 0.5225, loss: 1.5360, grad_norm: 8.0355
2023-04-23 16:38:42,793 - mmdet - INFO - Epoch [1][200/4399] lr: 7.972e-03, eta: 11:13:12, time: 0.665, data_time: 0.006, memory: 7504, loss_rpn_cls: 0.2514, loss_rpn_bbox: 0.2163, loss_cls: 0.4500, acc: 86.1561, loss_bbox: 0.5144, loss: 1.4320, grad_norm: 8.1893
2023-04-23 16:39:16,145 - mmdet - INFO - Epoch [1][250/4399] lr: 9.970e-03, eta: 10:54:52, time: 0.667, data_time: 0.006, memory: 7504, loss_rpn_cls: 0.2354, loss_rpn_bbox: 0.2269, loss_cls: 0.4661, acc: 85.6025, loss_bbox: 0.5505, loss: 1.4789, grad_norm: 8.3150

[Suncheng2022]

The results obtained from the implementation of end-to end inference are as follows，is it normal, the AP maybe low?:

Evaluating bbox...
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=300 catIds=all] = 0.139
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=300 catIds=all] = 0.216
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=300 catIds=all] = 0.151
Average Precision (AP) @[ IoU=0.50:0.95 | area= s | maxDets=300 catIds=all] = 0.077
Average Precision (AP) @[ IoU=0.50:0.95 | area= m | maxDets=300 catIds=all] = 0.167
Average Precision (AP) @[ IoU=0.50:0.95 | area= l | maxDets=300 catIds=all] = 0.291
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=300 catIds= r] = 0.176
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=300 catIds= c] = 0.137
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=300 catIds= f] = 0.128
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=300 catIds=all] = 0.197
Average Recall (AR) @[ IoU=0.50:0.95 | area= s | maxDets=300 catIds=all] = 0.095
Average Recall (AR) @[ IoU=0.50:0.95 | area= m | maxDets=300 catIds=all] = 0.217
Average Recall (AR) @[ IoU=0.50:0.95 | area= l | maxDets=300 catIds=all] = 0.395
OrderedDict([('bbox_AP', 0.139), ('bbox_AP50', 0.216), ('bbox_AP75', 0.151), ('bbox_APs', 0.077), ('bbox_APm', 0.167), ('bbox_APl', 0.291), ('bbox_APr', 0.176), ('bbox_APc', 0.137), ('bbox_APf', 0.128), ('bbox_mAP_copypaste', 'AP:0.139 AP50:0.216 AP75:0.151 APs:0.077 APm:0.167 APl:0.291 APr:0.176 APc:0.137 APf:0.128')])

[Rzx520]

Hello, may I ask how you obtained the lvis_v0.5 dataset，the official website only has v1

[Suncheng2022]

Hello, may I ask how you obtained the lvis_v0.5 dataset，the official website only has v1

Hello, I just search it in google. Maybe this page help.
(PS: I have been following on this project for a long time, but I have not been able to successfully run the new category evaluation and detection effect, if you are successful can you reply? Wish you success!)

[GitDu6]

@Suncheng2022 Hello,may I ask you why the end to end inference running so slowly ?