Skip to content

Commit

Permalink
Fix typo (open-mmlab#10976)
Browse files Browse the repository at this point in the history
  • Loading branch information
@hhaAndroid @yumion
hhaAndroid authored and yumion committed Jan 31, 2024
1 parent c88b75a commit 7b9e0ba
Show file tree
Hide file tree
Showing 2 changed files with 11 additions and 11 deletions.
14 changes: 7 additions & 7 deletions configs/glip/README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -58,23 +58,23 @@ tokenizer.save_pretrained("your path/bert-base-uncased")

## Results and Models

| Model | Zero-shot or Funetune | COCO mAP | Official COCO mAP | Pre-Train Data | Config | Download |
| Model | Zero-shot or Finetune | COCO mAP | Official COCO mAP | Pre-Train Data | Config | Download |
| :--------: | :-------------------: | :------: | ----------------: | :------------------------: | :---------------------------------------------------------------------: | :-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| GLIP-T (A) | Zero-shot | 43.0 | 42.9 | O365 | [config](glip_atss_swin-t_a_fpn_dyhead_pretrain_obj365.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_tiny_a_mmdet-b3654169.pth) |
| GLIP-T (A) | Funetune | 53.3 | 52.9 | O365 | [config](glip_atss_swin-t_a_fpn_dyhead_16xb2_ms-2x_funtune_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_a_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_a_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230914_180419-e6addd96.pth)\| [log](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_a_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_a_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230914_180419.log.json) |
| GLIP-T (A) | Finetune | 53.3 | 52.9 | O365 | [config](glip_atss_swin-t_a_fpn_dyhead_16xb2_ms-2x_funtune_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_a_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_a_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230914_180419-e6addd96.pth)\| [log](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_a_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_a_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230914_180419.log.json) |
| GLIP-T (B) | Zero-shot | 44.9 | 44.9 | O365 | [config](glip_atss_swin-t_b_fpn_dyhead_pretrain_obj365.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_tiny_b_mmdet-6dfbd102.pth) |
| GLIP-T (B) | Funetune | 54.1 | 53.8 | O365 | [config](glip_atss_swin-t_b_fpn_dyhead_16xb2_ms-2x_funtune_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_b_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_b_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230916_163538-650323ba.pth)\| [log](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_b_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_b_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230916_163538.log.json) |
| GLIP-T (B) | Finetune | 54.1 | 53.8 | O365 | [config](glip_atss_swin-t_b_fpn_dyhead_16xb2_ms-2x_funtune_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_b_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_b_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230916_163538-650323ba.pth)\| [log](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_b_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_b_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230916_163538.log.json) |
| GLIP-T (C) | Zero-shot | 46.7 | 46.7 | O365,GoldG | [config](glip_atss_swin-t_c_fpn_dyhead_pretrain_obj365-goldg.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_tiny_c_mmdet-2fc427dd.pth) |
| GLIP-T (C) | Funetune | 55.2 | 55.1 | O365,GoldG | [config](glip_atss_swin-t_c_fpn_dyhead_16xb2_ms-2x_funtune_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_c_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_c_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230914_182935-4ba3fc3b.pth)\| [log](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_c_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_c_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230914_182935.log.json) |
| GLIP-T (C) | Finetune | 55.2 | 55.1 | O365,GoldG | [config](glip_atss_swin-t_c_fpn_dyhead_16xb2_ms-2x_funtune_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_c_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_c_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230914_182935-4ba3fc3b.pth)\| [log](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_c_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_c_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230914_182935.log.json) |
| GLIP-T | Zero-shot | 46.6 | 46.6 | O365,GoldG,CC3M,SBU | [config](glip_atss_swin-t_fpn_dyhead_pretrain_obj365-goldg-cc3m-sub.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_tiny_mmdet-c24ce662.pth) |
| GLIP-T | Funetune | 55.4 | 55.2 | O365,GoldG,CC3M,SBU | [config](glip_atss_swin-t_fpn_dyhead_16xb2_ms-2x_funtune_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230914_224410-ba97be24.pth)\| [log](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230914_224410.log.json) |
| GLIP-T | Finetune | 55.4 | 55.2 | O365,GoldG,CC3M,SBU | [config](glip_atss_swin-t_fpn_dyhead_16xb2_ms-2x_funtune_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230914_224410-ba97be24.pth)\| [log](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-t_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-t_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230914_224410.log.json) |
| GLIP-L | Zero-shot | 51.3 | 51.4 | FourODs,GoldG,CC3M+12M,SBU | [config](glip_atss_swin-l_fpn_dyhead_pretrain_mixeddata.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_l_mmdet-abfe026b.pth) |
| GLIP-L | Funetune | 59.4 | | FourODs,GoldG,CC3M+12M,SBU | [config](glip_atss_swin-l_fpn_dyhead_16xb2_ms-2x_funtune_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-l_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-l_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230910_100800-e9be4274.pth)\| [log](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-l_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-l_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230910_100800.log.json) |
| GLIP-L | Finetune | 59.4 | | FourODs,GoldG,CC3M+12M,SBU | [config](glip_atss_swin-l_fpn_dyhead_16xb2_ms-2x_funtune_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-l_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-l_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230910_100800-e9be4274.pth)\| [log](https://download.openmmlab.com/mmdetection/v3.0/glip/glip_atss_swin-l_fpn_dyhead_16xb2_ms-2x_funtune_coco/glip_atss_swin-l_fpn_dyhead_16xb2_ms-2x_funtune_coco_20230910_100800.log.json) |

Note:

1. The weights corresponding to the zero-shot model are adopted from the official weights and converted using the [script](../../tools/model_converters/glip_to_mmdet.py). We have not retrained the model for the time being.
2. Funetune refers to fine-tuning on the COCO 2017 dataset. The L model is trained using 16 A100 GPUs, while the remaining models are trained using 16 NVIDIA GeForce 3090 GPUs.
2. Finetune refers to fine-tuning on the COCO 2017 dataset. The L model is trained using 16 A100 GPUs, while the remaining models are trained using 16 NVIDIA GeForce 3090 GPUs.
3. Taking the GLIP-T(A) model as an example, I trained it twice using the official code, and the fine-tuning mAP were 52.5 and 52.6. Therefore, the mAP we achieved in our reproduction is higher than the official results. The main reason is that we modified the `weight_decay` parameter.
4. Our experiments revealed that training for 24 epochs leads to overfitting. Therefore, we chose the best-performing model. If users want to train on a custom dataset, it is advisable to shorten the number of epochs and save the best-performing model.
5. Due to the official absence of fine-tuning hyperparameters for the GLIP-L model, we have not yet reproduced the official accuracy. I have found that overfitting can also occur, so it may be necessary to consider custom modifications to data augmentation and model enhancement. Given the high cost of training, we have not conducted any research on this matter at the moment.
8 changes: 4 additions & 4 deletions configs/grounding_dino/README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -64,13 +64,13 @@ python demo/image_demo.py \
| Model | Backbone | Style | COCO mAP | Official COCO mAP | Pre-Train Data | Config | Download |
| :----------------: | :------: | :-------: | :--------: | :---------------: | :----------------------------------------------: | :------------------------------------------------------------: | :---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------: |
| Grounding DINO-T | Swin-T | Zero-shot | 48.5 | 48.4 | O365,GoldG,Cap4M | [config](grounding_dino_swin-t_pretrain_obj365_goldg_cap4m.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/groundingdino_swint_ogc_mmdet-822d7e9d.pth) |
| Grounding DINO-T | Swin-T | Funetune | 58.1(+0.9) | 57.2 | O365,GoldG,Cap4M | [config](grounding_dino_swin-t_finetune_16xb2_1x_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/grounding_dino_swin-t_finetune_16xb2_1x_coco/grounding_dino_swin-t_finetune_16xb2_1x_coco_20230921_152544-5f234b20.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/grounding_dino_swin-t_finetune_16xb2_1x_coco/grounding_dino_swin-t_finetune_16xb2_1x_coco_20230921_152544.log.json) |
| Grounding DINO-T | Swin-T | Finetune | 58.1(+0.9) | 57.2 | O365,GoldG,Cap4M | [config](grounding_dino_swin-t_finetune_16xb2_1x_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/grounding_dino_swin-t_finetune_16xb2_1x_coco/grounding_dino_swin-t_finetune_16xb2_1x_coco_20230921_152544-5f234b20.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/grounding_dino_swin-t_finetune_16xb2_1x_coco/grounding_dino_swin-t_finetune_16xb2_1x_coco_20230921_152544.log.json) |
| Grounding DINO-B | Swin-B | Zero-shot | 56.9 | 56.7 | COCO,O365,GoldG,Cap4M,OpenImage,ODinW-35,RefCOCO | [config](grounding_dino_swin-b_pretrain_mixeddata.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/groundingdino_swinb_cogcoor_mmdet-55949c9c.pth) |
| Grounding DINO-B | Swin-B | Funetune | 59.7 | | COCO,O365,GoldG,Cap4M,OpenImage,ODinW-35,RefCOCO | [config](grounding_dino_swin-b_finetune_16xb2_1x_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/grounding_dino_swin-b_finetune_16xb2_1x_coco/grounding_dino_swin-b_finetune_16xb2_1x_coco_20230921_153201-f219e0c0.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/grounding_dino_swin-b_finetune_16xb2_1x_coco/grounding_dino_swin-b_finetune_16xb2_1x_coco_20230921_153201.log.json) |
| Grounding DINO-R50 | R50 | scratch | 48.9(+0.8) | 48.1 | | [config](grounding_dino_r50_scratch_8xb2_1x_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/grounding_dino_r50_scratch_8xb2_1x_coco/grounding_dino_r50_scratch_1x_coco-fe0002f2.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/grounding_dino_r50_scratch_8xb2_1x_coco/20230922_114218.json) |
| Grounding DINO-B | Swin-B | Finetune | 59.7 | | COCO,O365,GoldG,Cap4M,OpenImage,ODinW-35,RefCOCO | [config](grounding_dino_swin-b_finetune_16xb2_1x_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/grounding_dino_swin-b_finetune_16xb2_1x_coco/grounding_dino_swin-b_finetune_16xb2_1x_coco_20230921_153201-f219e0c0.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/grounding_dino_swin-b_finetune_16xb2_1x_coco/grounding_dino_swin-b_finetune_16xb2_1x_coco_20230921_153201.log.json) |
| Grounding DINO-R50 | R50 | Scratch | 48.9(+0.8) | 48.1 | | [config](grounding_dino_r50_scratch_8xb2_1x_coco.py) | [model](https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/grounding_dino_r50_scratch_8xb2_1x_coco/grounding_dino_r50_scratch_1x_coco-fe0002f2.pth) \| [log](https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/grounding_dino_r50_scratch_8xb2_1x_coco/20230922_114218.json) |

Note:

1. The weights corresponding to the zero-shot model are adopted from the official weights and converted using the [script](../../tools/model_converters/groundingdino_to_mmdet.py). We have not retrained the model for the time being.
2. Funetune refers to fine-tuning on the COCO 2017 dataset. The R50 model is trained using 8 NVIDIA GeForce 3090 GPUs, while the remaining models are trained using 16 NVIDIA GeForce 3090 GPUs. The GPU memory usage is approximately 8.5GB.
2. Finetune refers to fine-tuning on the COCO 2017 dataset. The R50 model is trained using 8 NVIDIA GeForce 3090 GPUs, while the remaining models are trained using 16 NVIDIA GeForce 3090 GPUs. The GPU memory usage is approximately 8.5GB.
3. Our performance is higher than the official model due to two reasons: we modified the initialization strategy and introduced a log scaler.

0 comments on commit 7b9e0ba

Please sign in to comment.