Confusion between the results on DIOR-RSVG dataset

[jian-rookie]

Thanks for your nice and open-souce work.
I utilize the released weights to infernce on DIOR-RSVG and RSVG-HG datasets. The results on RSVG-HG dataset are same with the paper.
But the results on DIOR-RSVG dataset are much higher than the reported results in the paper, which are shown as follows.

Inference only supports for batch size = 1
number of params: 41508886
Unexpected Keys: ['text_encoder.embeddings.position_ids']
[0/7500] Time 0.963 acc@0.5: 1.0000 acc@0.6: 1.0000 acc@0.7: 1.0000 acc@0.8: 1.0000 acc@0.9: 1.0000 meanIoU: 0.9701 cumuIoU: 0.9701
[50/7500] Time 0.076 acc@0.5: 0.8824 acc@0.6: 0.8824 acc@0.7: 0.8824 acc@0.8: 0.8235 acc@0.9: 0.7451 meanIoU: 0.8337 cumuIoU: 0.9216
[100/7500] Time 0.068 acc@0.5: 0.9406 acc@0.6: 0.9406 acc@0.7: 0.9406 acc@0.8: 0.8812 acc@0.9: 0.6931 meanIoU: 0.8737 cumuIoU: 0.9388
[150/7500] Time 0.065 acc@0.5: 0.9338 acc@0.6: 0.9272 acc@0.7: 0.9272 acc@0.8: 0.8808 acc@0.9: 0.6954 meanIoU: 0.8699 cumuIoU: 0.9223
[200/7500] Time 0.064 acc@0.5: 0.9353 acc@0.6: 0.9303 acc@0.7: 0.9254 acc@0.8: 0.8806 acc@0.9: 0.6866 meanIoU: 0.8700 cumuIoU: 0.9242
[250/7500] Time 0.063 acc@0.5: 0.9442 acc@0.6: 0.9402 acc@0.7: 0.9323 acc@0.8: 0.8845 acc@0.9: 0.7012 meanIoU: 0.8783 cumuIoU: 0.9280
[300/7500] Time 0.063 acc@0.5: 0.9535 acc@0.6: 0.9502 acc@0.7: 0.9369 acc@0.8: 0.8837 acc@0.9: 0.6910 meanIoU: 0.8839 cumuIoU: 0.9258
[350/7500] Time 0.062 acc@0.5: 0.9516 acc@0.6: 0.9487 acc@0.7: 0.9373 acc@0.8: 0.8889 acc@0.9: 0.6866 meanIoU: 0.8826 cumuIoU: 0.9228
[400/7500] Time 0.062 acc@0.5: 0.9526 acc@0.6: 0.9476 acc@0.7: 0.9377 acc@0.8: 0.8878 acc@0.9: 0.6883 meanIoU: 0.8820 cumuIoU: 0.9290
[450/7500] Time 0.062 acc@0.5: 0.9579 acc@0.6: 0.9534 acc@0.7: 0.9401 acc@0.8: 0.8914 acc@0.9: 0.7007 meanIoU: 0.8867 cumuIoU: 0.9308
[500/7500] Time 0.062 acc@0.5: 0.9561 acc@0.6: 0.9481 acc@0.7: 0.9341 acc@0.8: 0.8862 acc@0.9: 0.6906 meanIoU: 0.8828 cumuIoU: 0.9322
[550/7500] Time 0.061 acc@0.5: 0.9546 acc@0.6: 0.9474 acc@0.7: 0.9347 acc@0.8: 0.8857 acc@0.9: 0.6878 meanIoU: 0.8816 cumuIoU: 0.9325
[600/7500] Time 0.061 acc@0.5: 0.9567 acc@0.6: 0.9501 acc@0.7: 0.9334 acc@0.8: 0.8835 acc@0.9: 0.6872 meanIoU: 0.8824 cumuIoU: 0.9343
[650/7500] Time 0.061 acc@0.5: 0.9555 acc@0.6: 0.9478 acc@0.7: 0.9309 acc@0.8: 0.8817 acc@0.9: 0.6805 meanIoU: 0.8803 cumuIoU: 0.9356
[700/7500] Time 0.061 acc@0.5: 0.9586 acc@0.6: 0.9515 acc@0.7: 0.9330 acc@0.8: 0.8845 acc@0.9: 0.6847 meanIoU: 0.8830 cumuIoU: 0.9372
[750/7500] Time 0.061 acc@0.5: 0.9547 acc@0.6: 0.9467 acc@0.7: 0.9281 acc@0.8: 0.8788 acc@0.9: 0.6831 meanIoU: 0.8789 cumuIoU: 0.9358
[800/7500] Time 0.061 acc@0.5: 0.9551 acc@0.6: 0.9476 acc@0.7: 0.9276 acc@0.8: 0.8789 acc@0.9: 0.6792 meanIoU: 0.8791 cumuIoU: 0.9370
[850/7500] Time 0.061 acc@0.5: 0.9542 acc@0.6: 0.9459 acc@0.7: 0.9271 acc@0.8: 0.8813 acc@0.9: 0.6816 meanIoU: 0.8789 cumuIoU: 0.9380
[900/7500] Time 0.061 acc@0.5: 0.9534 acc@0.6: 0.9456 acc@0.7: 0.9256 acc@0.8: 0.8768 acc@0.9: 0.6792 meanIoU: 0.8779 cumuIoU: 0.9383
[950/7500] Time 0.061 acc@0.5: 0.9527 acc@0.6: 0.9443 acc@0.7: 0.9243 acc@0.8: 0.8749 acc@0.9: 0.6761 meanIoU: 0.8775 cumuIoU: 0.9390
[1000/7500] Time 0.061 acc@0.5: 0.9520 acc@0.6: 0.9441 acc@0.7: 0.9251 acc@0.8: 0.8771 acc@0.9: 0.6763 meanIoU: 0.8780 cumuIoU: 0.9386
[1050/7500] Time 0.061 acc@0.5: 0.9505 acc@0.6: 0.9429 acc@0.7: 0.9248 acc@0.8: 0.8763 acc@0.9: 0.6736 meanIoU: 0.8762 cumuIoU: 0.9352
[1100/7500] Time 0.061 acc@0.5: 0.9519 acc@0.6: 0.9437 acc@0.7: 0.9246 acc@0.8: 0.8765 acc@0.9: 0.6721 meanIoU: 0.8767 cumuIoU: 0.9349
[1150/7500] Time 0.061 acc@0.5: 0.9522 acc@0.6: 0.9435 acc@0.7: 0.9253 acc@0.8: 0.8784 acc@0.9: 0.6768 meanIoU: 0.8772 cumuIoU: 0.9358
[1200/7500] Time 0.061 acc@0.5: 0.9517 acc@0.6: 0.9434 acc@0.7: 0.9242 acc@0.8: 0.8734 acc@0.9: 0.6744 meanIoU: 0.8760 cumuIoU: 0.9351
[1250/7500] Time 0.061 acc@0.5: 0.9496 acc@0.6: 0.9416 acc@0.7: 0.9225 acc@0.8: 0.8729 acc@0.9: 0.6707 meanIoU: 0.8742 cumuIoU: 0.9322
[1300/7500] Time 0.061 acc@0.5: 0.9485 acc@0.6: 0.9408 acc@0.7: 0.9216 acc@0.8: 0.8732 acc@0.9: 0.6726 meanIoU: 0.8735 cumuIoU: 0.9324
[1350/7500] Time 0.061 acc@0.5: 0.9482 acc@0.6: 0.9408 acc@0.7: 0.9223 acc@0.8: 0.8734 acc@0.9: 0.6706 meanIoU: 0.8731 cumuIoU: 0.9328
[1400/7500] Time 0.061 acc@0.5: 0.9479 acc@0.6: 0.9393 acc@0.7: 0.9201 acc@0.8: 0.8722 acc@0.9: 0.6724 meanIoU: 0.8726 cumuIoU: 0.9312
[1450/7500] Time 0.061 acc@0.5: 0.9476 acc@0.6: 0.9394 acc@0.7: 0.9194 acc@0.8: 0.8725 acc@0.9: 0.6706 meanIoU: 0.8721 cumuIoU: 0.9311
[1500/7500] Time 0.061 acc@0.5: 0.9474 acc@0.6: 0.9394 acc@0.7: 0.9194 acc@0.8: 0.8734 acc@0.9: 0.6729 meanIoU: 0.8724 cumuIoU: 0.9305
[1550/7500] Time 0.061 acc@0.5: 0.9471 acc@0.6: 0.9387 acc@0.7: 0.9188 acc@0.8: 0.8723 acc@0.9: 0.6725 meanIoU: 0.8722 cumuIoU: 0.9306
[1600/7500] Time 0.060 acc@0.5: 0.9450 acc@0.6: 0.9369 acc@0.7: 0.9169 acc@0.8: 0.8688 acc@0.9: 0.6702 meanIoU: 0.8701 cumuIoU: 0.9253
[1650/7500] Time 0.060 acc@0.5: 0.9449 acc@0.6: 0.9370 acc@0.7: 0.9158 acc@0.8: 0.8686 acc@0.9: 0.6687 meanIoU: 0.8699 cumuIoU: 0.9248
[1700/7500] Time 0.060 acc@0.5: 0.9459 acc@0.6: 0.9377 acc@0.7: 0.9165 acc@0.8: 0.8689 acc@0.9: 0.6696 meanIoU: 0.8708 cumuIoU: 0.9250
[1750/7500] Time 0.060 acc@0.5: 0.9469 acc@0.6: 0.9389 acc@0.7: 0.9178 acc@0.8: 0.8692 acc@0.9: 0.6699 meanIoU: 0.8716 cumuIoU: 0.9261
[1800/7500] Time 0.060 acc@0.5: 0.9456 acc@0.6: 0.9373 acc@0.7: 0.9167 acc@0.8: 0.8690 acc@0.9: 0.6724 meanIoU: 0.8707 cumuIoU: 0.9248
[1850/7500] Time 0.060 acc@0.5: 0.9460 acc@0.6: 0.9379 acc@0.7: 0.9168 acc@0.8: 0.8682 acc@0.9: 0.6699 meanIoU: 0.8710 cumuIoU: 0.9256
[1900/7500] Time 0.060 acc@0.5: 0.9463 acc@0.6: 0.9385 acc@0.7: 0.9164 acc@0.8: 0.8685 acc@0.9: 0.6739 meanIoU: 0.8715 cumuIoU: 0.9258
[1950/7500] Time 0.060 acc@0.5: 0.9457 acc@0.6: 0.9380 acc@0.7: 0.9149 acc@0.8: 0.8672 acc@0.9: 0.6720 meanIoU: 0.8707 cumuIoU: 0.9260
[2000/7500] Time 0.060 acc@0.5: 0.9470 acc@0.6: 0.9395 acc@0.7: 0.9170 acc@0.8: 0.8681 acc@0.9: 0.6737 meanIoU: 0.8718 cumuIoU: 0.9265
[2050/7500] Time 0.060 acc@0.5: 0.9469 acc@0.6: 0.9395 acc@0.7: 0.9161 acc@0.8: 0.8669 acc@0.9: 0.6724 meanIoU: 0.8711 cumuIoU: 0.9257
[2100/7500] Time 0.060 acc@0.5: 0.9462 acc@0.6: 0.9391 acc@0.7: 0.9158 acc@0.8: 0.8677 acc@0.9: 0.6716 meanIoU: 0.8706 cumuIoU: 0.9262
[2150/7500] Time 0.060 acc@0.5: 0.9470 acc@0.6: 0.9400 acc@0.7: 0.9168 acc@0.8: 0.8684 acc@0.9: 0.6709 meanIoU: 0.8712 cumuIoU: 0.9261
[2200/7500] Time 0.060 acc@0.5: 0.9468 acc@0.6: 0.9400 acc@0.7: 0.9159 acc@0.8: 0.8678 acc@0.9: 0.6720 meanIoU: 0.8710 cumuIoU: 0.9261
[2250/7500] Time 0.060 acc@0.5: 0.9471 acc@0.6: 0.9400 acc@0.7: 0.9156 acc@0.8: 0.8672 acc@0.9: 0.6730 meanIoU: 0.8712 cumuIoU: 0.9256
[2300/7500] Time 0.060 acc@0.5: 0.9474 acc@0.6: 0.9400 acc@0.7: 0.9157 acc@0.8: 0.8670 acc@0.9: 0.6723 meanIoU: 0.8712 cumuIoU: 0.9259
[2350/7500] Time 0.060 acc@0.5: 0.9477 acc@0.6: 0.9405 acc@0.7: 0.9158 acc@0.8: 0.8669 acc@0.9: 0.6721 meanIoU: 0.8716 cumuIoU: 0.9259
[2400/7500] Time 0.060 acc@0.5: 0.9488 acc@0.6: 0.9413 acc@0.7: 0.9163 acc@0.8: 0.8676 acc@0.9: 0.6735 meanIoU: 0.8724 cumuIoU: 0.9263
[2450/7500] Time 0.060 acc@0.5: 0.9490 acc@0.6: 0.9412 acc@0.7: 0.9164 acc@0.8: 0.8666 acc@0.9: 0.6724 meanIoU: 0.8722 cumuIoU: 0.9266
[2500/7500] Time 0.060 acc@0.5: 0.9488 acc@0.6: 0.9408 acc@0.7: 0.9164 acc@0.8: 0.8661 acc@0.9: 0.6709 meanIoU: 0.8719 cumuIoU: 0.9273
[2550/7500] Time 0.060 acc@0.5: 0.9486 acc@0.6: 0.9408 acc@0.7: 0.9165 acc@0.8: 0.8667 acc@0.9: 0.6695 meanIoU: 0.8720 cumuIoU: 0.9276
[2600/7500] Time 0.060 acc@0.5: 0.9493 acc@0.6: 0.9416 acc@0.7: 0.9177 acc@0.8: 0.8685 acc@0.9: 0.6717 meanIoU: 0.8728 cumuIoU: 0.9259
[2650/7500] Time 0.060 acc@0.5: 0.9491 acc@0.6: 0.9412 acc@0.7: 0.9170 acc@0.8: 0.8668 acc@0.9: 0.6722 meanIoU: 0.8726 cumuIoU: 0.9266
[2700/7500] Time 0.060 acc@0.5: 0.9485 acc@0.6: 0.9404 acc@0.7: 0.9163 acc@0.8: 0.8663 acc@0.9: 0.6731 meanIoU: 0.8721 cumuIoU: 0.9268
[2750/7500] Time 0.060 acc@0.5: 0.9484 acc@0.6: 0.9404 acc@0.7: 0.9160 acc@0.8: 0.8648 acc@0.9: 0.6718 meanIoU: 0.8716 cumuIoU: 0.9266
[2800/7500] Time 0.060 acc@0.5: 0.9489 acc@0.6: 0.9411 acc@0.7: 0.9168 acc@0.8: 0.8665 acc@0.9: 0.6723 meanIoU: 0.8724 cumuIoU: 0.9269
[2850/7500] Time 0.060 acc@0.5: 0.9495 acc@0.6: 0.9418 acc@0.7: 0.9172 acc@0.8: 0.8667 acc@0.9: 0.6720 meanIoU: 0.8728 cumuIoU: 0.9273
[2900/7500] Time 0.060 acc@0.5: 0.9493 acc@0.6: 0.9417 acc@0.7: 0.9169 acc@0.8: 0.8659 acc@0.9: 0.6711 meanIoU: 0.8724 cumuIoU: 0.9277
[2950/7500] Time 0.060 acc@0.5: 0.9485 acc@0.6: 0.9410 acc@0.7: 0.9163 acc@0.8: 0.8655 acc@0.9: 0.6720 meanIoU: 0.8722 cumuIoU: 0.9284
[3000/7500] Time 0.060 acc@0.5: 0.9470 acc@0.6: 0.9394 acc@0.7: 0.9147 acc@0.8: 0.8627 acc@0.9: 0.6698 meanIoU: 0.8706 cumuIoU: 0.9263
[3050/7500] Time 0.060 acc@0.5: 0.9466 acc@0.6: 0.9390 acc@0.7: 0.9145 acc@0.8: 0.8627 acc@0.9: 0.6719 meanIoU: 0.8706 cumuIoU: 0.9250
[3100/7500] Time 0.060 acc@0.5: 0.9468 acc@0.6: 0.9391 acc@0.7: 0.9149 acc@0.8: 0.8636 acc@0.9: 0.6727 meanIoU: 0.8710 cumuIoU: 0.9252
[3150/7500] Time 0.060 acc@0.5: 0.9464 acc@0.6: 0.9387 acc@0.7: 0.9143 acc@0.8: 0.8629 acc@0.9: 0.6728 meanIoU: 0.8704 cumuIoU: 0.9259
[3200/7500] Time 0.060 acc@0.5: 0.9472 acc@0.6: 0.9397 acc@0.7: 0.9153 acc@0.8: 0.8632 acc@0.9: 0.6735 meanIoU: 0.8711 cumuIoU: 0.9261
[3250/7500] Time 0.060 acc@0.5: 0.9468 acc@0.6: 0.9394 acc@0.7: 0.9151 acc@0.8: 0.8628 acc@0.9: 0.6746 meanIoU: 0.8711 cumuIoU: 0.9262
[3300/7500] Time 0.060 acc@0.5: 0.9464 acc@0.6: 0.9388 acc@0.7: 0.9146 acc@0.8: 0.8628 acc@0.9: 0.6746 meanIoU: 0.8707 cumuIoU: 0.9266
[3350/7500] Time 0.060 acc@0.5: 0.9463 acc@0.6: 0.9385 acc@0.7: 0.9144 acc@0.8: 0.8630 acc@0.9: 0.6741 meanIoU: 0.8706 cumuIoU: 0.9272
[3400/7500] Time 0.060 acc@0.5: 0.9465 acc@0.6: 0.9383 acc@0.7: 0.9136 acc@0.8: 0.8624 acc@0.9: 0.6739 meanIoU: 0.8707 cumuIoU: 0.9273
[3450/7500] Time 0.060 acc@0.5: 0.9452 acc@0.6: 0.9365 acc@0.7: 0.9116 acc@0.8: 0.8612 acc@0.9: 0.6734 meanIoU: 0.8696 cumuIoU: 0.9274
[3500/7500] Time 0.060 acc@0.5: 0.9449 acc@0.6: 0.9363 acc@0.7: 0.9112 acc@0.8: 0.8603 acc@0.9: 0.6735 meanIoU: 0.8692 cumuIoU: 0.9275
[3550/7500] Time 0.060 acc@0.5: 0.9448 acc@0.6: 0.9364 acc@0.7: 0.9110 acc@0.8: 0.8600 acc@0.9: 0.6733 meanIoU: 0.8690 cumuIoU: 0.9271
[3600/7500] Time 0.060 acc@0.5: 0.9453 acc@0.6: 0.9370 acc@0.7: 0.9117 acc@0.8: 0.8609 acc@0.9: 0.6740 meanIoU: 0.8695 cumuIoU: 0.9275
[3650/7500] Time 0.060 acc@0.5: 0.9452 acc@0.6: 0.9370 acc@0.7: 0.9115 acc@0.8: 0.8603 acc@0.9: 0.6743 meanIoU: 0.8695 cumuIoU: 0.9276
[3700/7500] Time 0.060 acc@0.5: 0.9451 acc@0.6: 0.9370 acc@0.7: 0.9119 acc@0.8: 0.8603 acc@0.9: 0.6741 meanIoU: 0.8694 cumuIoU: 0.9276
[3750/7500] Time 0.060 acc@0.5: 0.9451 acc@0.6: 0.9371 acc@0.7: 0.9118 acc@0.8: 0.8600 acc@0.9: 0.6745 meanIoU: 0.8694 cumuIoU: 0.9281
[3800/7500] Time 0.060 acc@0.5: 0.9445 acc@0.6: 0.9366 acc@0.7: 0.9116 acc@0.8: 0.8600 acc@0.9: 0.6748 meanIoU: 0.8690 cumuIoU: 0.9284
[3850/7500] Time 0.060 acc@0.5: 0.9452 acc@0.6: 0.9372 acc@0.7: 0.9125 acc@0.8: 0.8613 acc@0.9: 0.6767 meanIoU: 0.8698 cumuIoU: 0.9288
[3900/7500] Time 0.060 acc@0.5: 0.9449 acc@0.6: 0.9369 acc@0.7: 0.9118 acc@0.8: 0.8611 acc@0.9: 0.6765 meanIoU: 0.8696 cumuIoU: 0.9292
[3950/7500] Time 0.060 acc@0.5: 0.9451 acc@0.6: 0.9370 acc@0.7: 0.9119 acc@0.8: 0.8613 acc@0.9: 0.6770 meanIoU: 0.8698 cumuIoU: 0.9295
[4000/7500] Time 0.060 acc@0.5: 0.9448 acc@0.6: 0.9363 acc@0.7: 0.9115 acc@0.8: 0.8598 acc@0.9: 0.6761 meanIoU: 0.8693 cumuIoU: 0.9290
[4050/7500] Time 0.060 acc@0.5: 0.9445 acc@0.6: 0.9358 acc@0.7: 0.9111 acc@0.8: 0.8586 acc@0.9: 0.6754 meanIoU: 0.8691 cumuIoU: 0.9289
[4100/7500] Time 0.060 acc@0.5: 0.9449 acc@0.6: 0.9364 acc@0.7: 0.9117 acc@0.8: 0.8591 acc@0.9: 0.6747 meanIoU: 0.8694 cumuIoU: 0.9290
[4150/7500] Time 0.060 acc@0.5: 0.9453 acc@0.6: 0.9369 acc@0.7: 0.9123 acc@0.8: 0.8596 acc@0.9: 0.6760 meanIoU: 0.8699 cumuIoU: 0.9295
[4200/7500] Time 0.060 acc@0.5: 0.9448 acc@0.6: 0.9360 acc@0.7: 0.9114 acc@0.8: 0.8591 acc@0.9: 0.6770 meanIoU: 0.8695 cumuIoU: 0.9295
[4250/7500] Time 0.060 acc@0.5: 0.9447 acc@0.6: 0.9360 acc@0.7: 0.9113 acc@0.8: 0.8586 acc@0.9: 0.6773 meanIoU: 0.8695 cumuIoU: 0.9292
[4300/7500] Time 0.060 acc@0.5: 0.9447 acc@0.6: 0.9361 acc@0.7: 0.9116 acc@0.8: 0.8596 acc@0.9: 0.6784 meanIoU: 0.8697 cumuIoU: 0.9298
[4350/7500] Time 0.060 acc@0.5: 0.9446 acc@0.6: 0.9361 acc@0.7: 0.9117 acc@0.8: 0.8598 acc@0.9: 0.6785 meanIoU: 0.8697 cumuIoU: 0.9300
[4400/7500] Time 0.060 acc@0.5: 0.9443 acc@0.6: 0.9359 acc@0.7: 0.9118 acc@0.8: 0.8598 acc@0.9: 0.6789 meanIoU: 0.8697 cumuIoU: 0.9296
[4450/7500] Time 0.060 acc@0.5: 0.9445 acc@0.6: 0.9362 acc@0.7: 0.9122 acc@0.8: 0.8598 acc@0.9: 0.6780 meanIoU: 0.8696 cumuIoU: 0.9295
[4500/7500] Time 0.060 acc@0.5: 0.9449 acc@0.6: 0.9367 acc@0.7: 0.9127 acc@0.8: 0.8600 acc@0.9: 0.6776 meanIoU: 0.8700 cumuIoU: 0.9297
[4550/7500] Time 0.060 acc@0.5: 0.9446 acc@0.6: 0.9365 acc@0.7: 0.9125 acc@0.8: 0.8589 acc@0.9: 0.6759 meanIoU: 0.8697 cumuIoU: 0.9289
[4600/7500] Time 0.060 acc@0.5: 0.9446 acc@0.6: 0.9365 acc@0.7: 0.9128 acc@0.8: 0.8589 acc@0.9: 0.6753 meanIoU: 0.8698 cumuIoU: 0.9291
[4650/7500] Time 0.060 acc@0.5: 0.9450 acc@0.6: 0.9368 acc@0.7: 0.9134 acc@0.8: 0.8594 acc@0.9: 0.6762 meanIoU: 0.8702 cumuIoU: 0.9294
[4700/7500] Time 0.060 acc@0.5: 0.9451 acc@0.6: 0.9370 acc@0.7: 0.9134 acc@0.8: 0.8600 acc@0.9: 0.6767 meanIoU: 0.8704 cumuIoU: 0.9295
[4750/7500] Time 0.060 acc@0.5: 0.9451 acc@0.6: 0.9371 acc@0.7: 0.9137 acc@0.8: 0.8600 acc@0.9: 0.6778 meanIoU: 0.8705 cumuIoU: 0.9298
[4800/7500] Time 0.060 acc@0.5: 0.9450 acc@0.6: 0.9369 acc@0.7: 0.9136 acc@0.8: 0.8600 acc@0.9: 0.6767 meanIoU: 0.8703 cumuIoU: 0.9289
[4850/7500] Time 0.060 acc@0.5: 0.9452 acc@0.6: 0.9371 acc@0.7: 0.9140 acc@0.8: 0.8606 acc@0.9: 0.6772 meanIoU: 0.8705 cumuIoU: 0.9291
[4900/7500] Time 0.060 acc@0.5: 0.9455 acc@0.6: 0.9376 acc@0.7: 0.9145 acc@0.8: 0.8610 acc@0.9: 0.6784 meanIoU: 0.8708 cumuIoU: 0.9293
[4950/7500] Time 0.060 acc@0.5: 0.9455 acc@0.6: 0.9376 acc@0.7: 0.9144 acc@0.8: 0.8608 acc@0.9: 0.6776 meanIoU: 0.8706 cumuIoU: 0.9293
[5000/7500] Time 0.060 acc@0.5: 0.9456 acc@0.6: 0.9378 acc@0.7: 0.9146 acc@0.8: 0.8614 acc@0.9: 0.6783 meanIoU: 0.8708 cumuIoU: 0.9284
[5050/7500] Time 0.060 acc@0.5: 0.9454 acc@0.6: 0.9370 acc@0.7: 0.9139 acc@0.8: 0.8610 acc@0.9: 0.6777 meanIoU: 0.8703 cumuIoU: 0.9283
[5100/7500] Time 0.060 acc@0.5: 0.9453 acc@0.6: 0.9371 acc@0.7: 0.9139 acc@0.8: 0.8606 acc@0.9: 0.6785 meanIoU: 0.8703 cumuIoU: 0.9281
[5150/7500] Time 0.060 acc@0.5: 0.9456 acc@0.6: 0.9375 acc@0.7: 0.9146 acc@0.8: 0.8608 acc@0.9: 0.6785 meanIoU: 0.8705 cumuIoU: 0.9282
[5200/7500] Time 0.060 acc@0.5: 0.9452 acc@0.6: 0.9371 acc@0.7: 0.9139 acc@0.8: 0.8600 acc@0.9: 0.6776 meanIoU: 0.8701 cumuIoU: 0.9275
[5250/7500] Time 0.060 acc@0.5: 0.9452 acc@0.6: 0.9372 acc@0.7: 0.9137 acc@0.8: 0.8600 acc@0.9: 0.6774 meanIoU: 0.8701 cumuIoU: 0.9275
[5300/7500] Time 0.060 acc@0.5: 0.9451 acc@0.6: 0.9370 acc@0.7: 0.9138 acc@0.8: 0.8598 acc@0.9: 0.6784 meanIoU: 0.8702 cumuIoU: 0.9265
[5350/7500] Time 0.060 acc@0.5: 0.9449 acc@0.6: 0.9366 acc@0.7: 0.9135 acc@0.8: 0.8598 acc@0.9: 0.6784 meanIoU: 0.8699 cumuIoU: 0.9269
[5400/7500] Time 0.060 acc@0.5: 0.9450 acc@0.6: 0.9369 acc@0.7: 0.9135 acc@0.8: 0.8600 acc@0.9: 0.6797 meanIoU: 0.8701 cumuIoU: 0.9270
[5450/7500] Time 0.060 acc@0.5: 0.9448 acc@0.6: 0.9367 acc@0.7: 0.9136 acc@0.8: 0.8598 acc@0.9: 0.6795 meanIoU: 0.8700 cumuIoU: 0.9270
[5500/7500] Time 0.060 acc@0.5: 0.9446 acc@0.6: 0.9364 acc@0.7: 0.9131 acc@0.8: 0.8597 acc@0.9: 0.6791 meanIoU: 0.8697 cumuIoU: 0.9273
[5550/7500] Time 0.060 acc@0.5: 0.9442 acc@0.6: 0.9360 acc@0.7: 0.9126 acc@0.8: 0.8589 acc@0.9: 0.6788 meanIoU: 0.8692 cumuIoU: 0.9274
[5600/7500] Time 0.060 acc@0.5: 0.9445 acc@0.6: 0.9361 acc@0.7: 0.9127 acc@0.8: 0.8590 acc@0.9: 0.6783 meanIoU: 0.8693 cumuIoU: 0.9278
[5650/7500] Time 0.060 acc@0.5: 0.9444 acc@0.6: 0.9361 acc@0.7: 0.9129 acc@0.8: 0.8593 acc@0.9: 0.6786 meanIoU: 0.8694 cumuIoU: 0.9273
[5700/7500] Time 0.060 acc@0.5: 0.9442 acc@0.6: 0.9356 acc@0.7: 0.9121 acc@0.8: 0.8583 acc@0.9: 0.6774 meanIoU: 0.8689 cumuIoU: 0.9272
[5750/7500] Time 0.060 acc@0.5: 0.9435 acc@0.6: 0.9350 acc@0.7: 0.9111 acc@0.8: 0.8571 acc@0.9: 0.6766 meanIoU: 0.8682 cumuIoU: 0.9261
[5800/7500] Time 0.060 acc@0.5: 0.9435 acc@0.6: 0.9350 acc@0.7: 0.9112 acc@0.8: 0.8573 acc@0.9: 0.6764 meanIoU: 0.8682 cumuIoU: 0.9263
[5850/7500] Time 0.060 acc@0.5: 0.9433 acc@0.6: 0.9349 acc@0.7: 0.9113 acc@0.8: 0.8569 acc@0.9: 0.6763 meanIoU: 0.8682 cumuIoU: 0.9262
[5900/7500] Time 0.060 acc@0.5: 0.9434 acc@0.6: 0.9351 acc@0.7: 0.9117 acc@0.8: 0.8571 acc@0.9: 0.6768 meanIoU: 0.8684 cumuIoU: 0.9263
[5950/7500] Time 0.060 acc@0.5: 0.9435 acc@0.6: 0.9351 acc@0.7: 0.9116 acc@0.8: 0.8570 acc@0.9: 0.6764 meanIoU: 0.8684 cumuIoU: 0.9263
[6000/7500] Time 0.060 acc@0.5: 0.9435 acc@0.6: 0.9352 acc@0.7: 0.9117 acc@0.8: 0.8574 acc@0.9: 0.6771 meanIoU: 0.8686 cumuIoU: 0.9266
[6050/7500] Time 0.060 acc@0.5: 0.9436 acc@0.6: 0.9352 acc@0.7: 0.9119 acc@0.8: 0.8580 acc@0.9: 0.6777 meanIoU: 0.8688 cumuIoU: 0.9268
[6100/7500] Time 0.060 acc@0.5: 0.9433 acc@0.6: 0.9349 acc@0.7: 0.9117 acc@0.8: 0.8581 acc@0.9: 0.6774 meanIoU: 0.8685 cumuIoU: 0.9269
[6150/7500] Time 0.060 acc@0.5: 0.9429 acc@0.6: 0.9346 acc@0.7: 0.9111 acc@0.8: 0.8573 acc@0.9: 0.6762 meanIoU: 0.8680 cumuIoU: 0.9271
[6200/7500] Time 0.060 acc@0.5: 0.9424 acc@0.6: 0.9342 acc@0.7: 0.9103 acc@0.8: 0.8566 acc@0.9: 0.6759 meanIoU: 0.8676 cumuIoU: 0.9273
[6250/7500] Time 0.060 acc@0.5: 0.9429 acc@0.6: 0.9344 acc@0.7: 0.9101 acc@0.8: 0.8565 acc@0.9: 0.6757 meanIoU: 0.8678 cumuIoU: 0.9273
[6300/7500] Time 0.060 acc@0.5: 0.9430 acc@0.6: 0.9343 acc@0.7: 0.9099 acc@0.8: 0.8565 acc@0.9: 0.6762 meanIoU: 0.8679 cumuIoU: 0.9265
[6350/7500] Time 0.060 acc@0.5: 0.9433 acc@0.6: 0.9343 acc@0.7: 0.9098 acc@0.8: 0.8559 acc@0.9: 0.6760 meanIoU: 0.8678 cumuIoU: 0.9265
[6400/7500] Time 0.060 acc@0.5: 0.9434 acc@0.6: 0.9345 acc@0.7: 0.9102 acc@0.8: 0.8564 acc@0.9: 0.6769 meanIoU: 0.8681 cumuIoU: 0.9266
[6450/7500] Time 0.060 acc@0.5: 0.9436 acc@0.6: 0.9347 acc@0.7: 0.9102 acc@0.8: 0.8569 acc@0.9: 0.6779 meanIoU: 0.8684 cumuIoU: 0.9269
[6500/7500] Time 0.060 acc@0.5: 0.9434 acc@0.6: 0.9345 acc@0.7: 0.9100 acc@0.8: 0.8568 acc@0.9: 0.6779 meanIoU: 0.8682 cumuIoU: 0.9267
[6550/7500] Time 0.060 acc@0.5: 0.9432 acc@0.6: 0.9342 acc@0.7: 0.9099 acc@0.8: 0.8571 acc@0.9: 0.6784 meanIoU: 0.8682 cumuIoU: 0.9262
[6600/7500] Time 0.060 acc@0.5: 0.9430 acc@0.6: 0.9341 acc@0.7: 0.9099 acc@0.8: 0.8567 acc@0.9: 0.6779 meanIoU: 0.8680 cumuIoU: 0.9261
[6650/7500] Time 0.060 acc@0.5: 0.9429 acc@0.6: 0.9340 acc@0.7: 0.9095 acc@0.8: 0.8564 acc@0.9: 0.6781 meanIoU: 0.8679 cumuIoU: 0.9264
[6700/7500] Time 0.060 acc@0.5: 0.9430 acc@0.6: 0.9340 acc@0.7: 0.9096 acc@0.8: 0.8566 acc@0.9: 0.6778 meanIoU: 0.8680 cumuIoU: 0.9268
[6750/7500] Time 0.060 acc@0.5: 0.9433 acc@0.6: 0.9344 acc@0.7: 0.9101 acc@0.8: 0.8571 acc@0.9: 0.6781 meanIoU: 0.8683 cumuIoU: 0.9267
[6800/7500] Time 0.060 acc@0.5: 0.9434 acc@0.6: 0.9346 acc@0.7: 0.9103 acc@0.8: 0.8575 acc@0.9: 0.6787 meanIoU: 0.8684 cumuIoU: 0.9268
[6850/7500] Time 0.060 acc@0.5: 0.9434 acc@0.6: 0.9346 acc@0.7: 0.9105 acc@0.8: 0.8577 acc@0.9: 0.6789 meanIoU: 0.8683 cumuIoU: 0.9268
[6900/7500] Time 0.060 acc@0.5: 0.9433 acc@0.6: 0.9346 acc@0.7: 0.9106 acc@0.8: 0.8580 acc@0.9: 0.6796 meanIoU: 0.8684 cumuIoU: 0.9270
[6950/7500] Time 0.060 acc@0.5: 0.9436 acc@0.6: 0.9347 acc@0.7: 0.9107 acc@0.8: 0.8583 acc@0.9: 0.6798 meanIoU: 0.8685 cumuIoU: 0.9265
[7000/7500] Time 0.060 acc@0.5: 0.9436 acc@0.6: 0.9347 acc@0.7: 0.9107 acc@0.8: 0.8586 acc@0.9: 0.6799 meanIoU: 0.8686 cumuIoU: 0.9266
[7050/7500] Time 0.060 acc@0.5: 0.9438 acc@0.6: 0.9350 acc@0.7: 0.9112 acc@0.8: 0.8589 acc@0.9: 0.6799 meanIoU: 0.8689 cumuIoU: 0.9266
[7100/7500] Time 0.060 acc@0.5: 0.9441 acc@0.6: 0.9354 acc@0.7: 0.9114 acc@0.8: 0.8592 acc@0.9: 0.6812 meanIoU: 0.8692 cumuIoU: 0.9268
[7150/7500] Time 0.060 acc@0.5: 0.9442 acc@0.6: 0.9355 acc@0.7: 0.9116 acc@0.8: 0.8595 acc@0.9: 0.6820 meanIoU: 0.8693 cumuIoU: 0.9270
[7200/7500] Time 0.060 acc@0.5: 0.9440 acc@0.6: 0.9353 acc@0.7: 0.9114 acc@0.8: 0.8593 acc@0.9: 0.6820 meanIoU: 0.8692 cumuIoU: 0.9268
[7250/7500] Time 0.060 acc@0.5: 0.9444 acc@0.6: 0.9357 acc@0.7: 0.9120 acc@0.8: 0.8595 acc@0.9: 0.6820 meanIoU: 0.8694 cumuIoU: 0.9270
[7300/7500] Time 0.060 acc@0.5: 0.9445 acc@0.6: 0.9359 acc@0.7: 0.9122 acc@0.8: 0.8597 acc@0.9: 0.6820 meanIoU: 0.8696 cumuIoU: 0.9268
[7350/7500] Time 0.060 acc@0.5: 0.9446 acc@0.6: 0.9361 acc@0.7: 0.9123 acc@0.8: 0.8599 acc@0.9: 0.6818 meanIoU: 0.8697 cumuIoU: 0.9270
[7400/7500] Time 0.060 acc@0.5: 0.9445 acc@0.6: 0.9360 acc@0.7: 0.9123 acc@0.8: 0.8599 acc@0.9: 0.6810 meanIoU: 0.8695 cumuIoU: 0.9264
[7450/7500] Time 0.060 acc@0.5: 0.9447 acc@0.6: 0.9361 acc@0.7: 0.9126 acc@0.8: 0.8606 acc@0.9: 0.6815 meanIoU: 0.8698 cumuIoU: 0.9266
acc@0.5: 0.9445 acc@0.6: 0.9360 acc@0.7: 0.9123 acc@0.8: 0.8601 acc@0.9: 0.6817 meanIoU: 0.8696 cumuIoU: 0.9266
test

I am confused about this difference. Hope for your reply.

[cxqNet]

I also encountered this problem. Have you found and solved it?

[xuancan1]

I also encountered this problem. Have you found and solved it?

请问这个问题解决了吗，而且我训练的模型效果74多

[jian-rookie]

I download the dataset from the DIOR-RSVG and re-train the LQVG. LQVG uses multi-scale training and multi-scale testing, but the original paper of DIOR-RSVG uses a fixed size of $640\times 640$ for training. So when I reproduce LQVG, I follow the single-scale training of the original paper. However, the single-scale inference code has some bugs, so multi-scale testing is still used.The reproduction results are as follows:
Pr@0.5: 78.71, Pr@0.6: 75.07, Pr@0.7: 69.21, Pr@0.8: 58.35, Pr@0.9: 33.99, mIoU: 68.82, cIoU: 78.36

[LANMNG]

I am retraining the model and will release the new model weights. Please wait a few days. Thanks.

[xuancan1]

我正在重新训练模型，并将发布新的模型权重。请稍等几天。谢谢。

感谢您出色的工作

[xuancan1]

我从 DIOR-RSVG 下载数据集并重新训练 LQVG。LQVG 采用多尺度训练和多尺度测试，但 DIOR-RSVG 的原论文使用固定大小 640 × 640 用于培训。所以当我复制 LQVG 时，我遵循原始论文的单尺度训练。但是，单尺度推理代码存在一些错误，因此仍然使用多尺度测试。繁殖结果为：Pr@0.5：78.71、Pr@0.6：75.07、Pr@0.7：69.21、Pr@0.8：58.35、Pr@0.9：33.99、mIoU：68.82、cIoU：78.36

所以我需要将数据处理中的图像尺寸固定为640×640，然后重新开始训练和测试，这样是不是能和论文的结果相近？

[jian-rookie]

我从 DIOR-RSVG 下载数据集并重新训练 LQVG。LQVG 采用多尺度训练和多尺度测试，但 DIOR-RSVG 的原论文使用固定大小 640 × 640 用于培训。所以当我复制 LQVG 时，我遵循原始论文的单尺度训练。但是，单尺度推理代码存在一些错误，因此仍然使用多尺度测试。繁殖结果为：Pr@0.5：78.71、Pr@0.6：75.07、Pr@0.7：69.21、Pr@0.8：58.35、Pr@0.9：33.99、mIoU：68.82、cIoU：78.36

所以我需要将数据处理中的图像尺寸固定为640×640，然后重新开始训练和测试，这样是不是能和论文的结果相近？

I choose single-scale training to make a relatively fair comparison with the results of the original paper of DIOR-RSVG. In theory, if you want to get an experimental result comparable to the paper of LQVG, you should implement multi-scale training and testing mentioned in the paper of LQVG.

It is best to wait for the original author's retrained model weights.

[LANMNG]

Hi, I have released a new retrained model weight to replace the previous wrong model weight, and we get a very similar result to the paper as follows:
acc@0.5: 0.8277 acc@0.6: 0.8043 acc@0.7: 0.7536 acc@0.8: 0.6405 acc@0.9: 0.4176 meanIoU: 0.7320 cumuIoU: 0.8197.
I know that the result is slightly lower than the reported result in the paper, and I am still training the model to get the same performance as the paper. Also, we use multi-scale training and single-scale inference. Moreover, considering that the training environment is a little difficult to configure, we also release our conda environment, hoping to help you reproduce the experimental results.

[xuancan1]

Hi, I have released a new retrained model weight to replace the previous wrong model weight, and we get a very similar result to the paper as follows: acc@0.5: 0.8277 acc@0.6: 0.8043 acc@0.7: 0.7536 acc@0.8: 0.6405 acc@0.9: 0.4176 meanIoU: 0.7320 cumuIoU: 0.8197. I know that the result is slightly lower than the reported result in the paper, and I am still training the model to get the same performance as the paper. Also, we use multi-scale training and single-scale inference. Moreover, considering that the training environment is a little difficult to configure, we also release our conda environment, hoping to help you reproduce the experimental results.

Thanks for your excellent and open-souce work.

[ThMay19]

Thanks for the excellent work and author’s maintenance at first！

I have tested the author's retrained model weights. The results are as follows:
acc@0.5: 0.9441 acc@0.6: 0.9363 acc@0.7: 0.9144 acc@0.8: 0.8600 acc@0.9: 0.6848 meanIoU: 0.8705 cumuIoU: 0.9231

If I understand correctly, the DIOR-RSVG dataset uses the "filelist" function to retrieve all XML files and traverses to read the 'object' . Then during the loop process, the dataset loading strategy utilizes the Index from "train.txt" and "test.txt" to split the final training and testing objects by "count += 1".

However, in different machines, the os.walk() method in "filelist" may return a list of different file names, ultimately arising in inconsistent dataset partitioning.

I'm not sure if I understand the loading strategy of the DIOR-RSVG dataset correctly. If so, that means the dataset split differs in different machines, which will arise unfair comparisons.

I really would like know how to test the model correctly in this case.

[halen-code]

Thanks for your excellent work.
Recently, I have a job that requires comparison with LQVG. But I found that there are some anomalies in the indicators of the LQVG model.
I not only cannot reproduce the effect on DIOR-RSVG, but I also cannot reproduce the effect on rsvg-hr. As mentioned by others before, the performance of the replicated LQVG on DIOR-RSVG is much lower than that shown in the paper. And I downloaded the weights pre trained by the author on DIOR-RSVG, and then fine tuned the RSVG-HR dataset. The results are as follows:
acc@0.5 : 0.3246 acc@0.6 : 0.3046 acc@0.7 : 0.2625 acc@0.8 : 0.1824 acc@0.9 : 0.0621 meanIoU: 0.2754 cumuIoU: 0.1874
This is far below the data of 87 in the paper. And as others have responded, the weights trained by the author on DIOR-RSVG performed exceptionally well on their test set, reaching 94, far exceeding the paper's metric of 83. Therefore, I suspect that there is an issue with the author's dataset. Did the model use the test set during training? Did you cheat? How should this be explained?