megvii-research/AnchorDETR

Issues

• 载入预训练模型发现attention 维度不对，可能的问题是 ？

  #50 opened 10 months ago by yelj
  4

• Multiple Predictions for Each Anchor Point

  #49 opened a year ago by zhangpeng2001
  1

• How to train one's own dataset

  #48 opened a year ago by LL-XSJ
  8

• Question about calculating GFLOPs

  #47 opened 2 years ago by yao-zheng-yi
  2

• Is there any way to get the coordinates(polygon shape returns)?

  #46 opened 2 years ago by yellowjs0304
  4

• How to replace resnet50 with your own backbone

  #45 opened 2 years ago by yao-zheng-yi
  2

• bs, l, c, h, w = srcs.shape

  #44 opened 2 years ago by feimadada
  1

• Why GroupNorm in self.input_proj ?

  #43 opened 2 years ago by liming-ai
  2

• Some questions about reference point and object query?

  #42 opened 2 years ago by rOtking
  6

• The comparison with DETR

  #39 opened 2 years ago by superwj1990
  4

• 测试问题

  #37 opened 2 years ago by jiangyichen19
  6

• Does the model always output 100 detection results on one image?

  #41 opened 2 years ago by ZM-Zhou
  2

• 关于feature shape的问题

  #40 opened 2 years ago by stillwaterman
  2

• Hello! The Feature position in figure3 is the positional encoding in DETR

  #36 opened 2 years ago by cxq1
  2

• Hello the default model is AnchorDETR-C5 or AnchorDETR-DC5 ?

  #34 opened 3 years ago by cxq1
  2

• Why not input updated reference points to next layer decoder?

  #35 opened 3 years ago by JarvisUSTC
  2

• Trying to understand my results on a custom dataset

  #29 opened 3 years ago by jackonealll
  9

• Details about GFLOPs

  #30 opened 3 years ago by volgachen
  3

• Points visualization

  #32 opened 3 years ago by Mark-Laohu
  5

• num_feature_levels > 1

  #9 opened 3 years ago by yformer
  43

• question about extended to panoptic segmentation

  #15 opened 3 years ago by luohao123
  21

• NAN error: assert (boxes1[:, 2:] >= boxes1[:, :2]).all()

  #16 opened 3 years ago by yformer
  10

• Model can not achieve a reasonable performance

  #20 opened 3 years ago by luohao123
  8

• About batch size and lr

  #21 opened 3 years ago by helq2612
  3

• Pretrained weights transfer to d2 evaluation result not same

  #23 opened 3 years ago by luohao123
  3

• something wrong with evaluating

  #25 opened 3 years ago by yihui8776
  2

• About Visualization

  #27 opened 3 years ago by zhouzheng1123
  3

• Code for PointDETR

  #33 opened 3 years ago by saratrajput
  1

• coco test-dev

  #17 opened 3 years ago by helq2612
  4

• points visualization

  #31 opened 3 years ago by Mark-Laohu
  0

• Load pretrained model AnchorDETR-DC5.

  #28 opened 3 years ago by NguyenVanThanhHust
  2

• How extract precision, recall and f1-score metrics

  #26 opened 3 years ago by jackonealll
  5

• Downloaded AnchorDETR R50 C5 weights evaluation only got 38.5

  #24 opened 3 years ago by lucasjinreal
  3

• How to get the 151 GFLOPs?

  #22 opened 3 years ago by charlesCXK
  2

• AP stops at 30 at iteration 250000

  #19 opened 3 years ago by luohao123
  1

• 关于在自己数据集上的应用

  #13 opened 3 years ago by Breeze-Zero
  1

• GPU memory consumption of AnchorDETR compared to DeformableDETR

  #14 opened 3 years ago by vadimkantorov
  2

• Overfitting when loading the pretrained model to train own dataset

  #12 opened 3 years ago by zknus
  15

• why the output is 900 while DETR is 100

  #11 opened 3 years ago by lucasjinreal
  3

• onnx model can not be simpiflied and pass onnx.check and wierd output

  #10 opened 3 years ago by lucasjinreal
  10

• How many encoder and decoder layers used in AnchorDETR?

  #3 opened 3 years ago by lucasjinreal
  30

• Pretrained model

  #5 opened 3 years ago by chenhbo
  3

• Your module has parameters that were not used in producing a loss.

  #6 opened 3 years ago by zongdaoming
  3

• Performance of the baseline

  #7 opened 3 years ago by ZhangGongjie
  1

• how to use custom data

  #8 opened 3 years ago by ffffwyy
  3

• Question about the time cost per epoch

  #4 opened 3 years ago by tymanman
  4

• How to finetune AnchorDETR on custom dataset?

  #1 opened 3 years ago by ducvuuit
  2

• Some questions about your code

  #2 opened 3 years ago by Huzhen757
  7