CenterNet/readme/MODEL_ZOO.md

89 lines
7.7 KiB
Markdown

# MODEL ZOO
### Common settings and notes
- The experiments are run with pytorch 0.4.1, CUDA 9.0, and CUDNN 7.1.
- Training times are measured on our servers with 8 TITAN V GPUs (12 GB Memeory).
- Testing times are measured on our local machine with TITAN Xp GPU.
- The models can be downloaded directly from [Google drive](https://drive.google.com/open?id=1px-Xg7jXSC79QqgsD1AAGJQkuf5m0zh_).
## Object Detection
### COCO
| Model | GPUs |Train time(h)| Test time (ms) | AP | Download |
|--------------------------|------|-------------|----------------|--------------------|-----------|
|[ctdet\_coco\_hg](../experiments/ctdet_coco_hg.sh) | 5 |109 | 71 / 129 / 674 | 40.3 / 42.2 / 45.1 | [model](https://drive.google.com/open?id=1cNyDmyorOduMRsgXoUnuyUiF6tZNFxaG) |
|[ctdet\_coco\_dla\_1x](../experiments/ctdet_coco_dla_1x.sh) | 8 | 57 | 19 / 36 / 248 | 36.3 / 38.2 / 40.7 | [model](https://drive.google.com/open?id=1r89_KNXyDyvUp8NggduG9uKQTMU2DsK_) |
|[ctdet\_coco\_dla\_2x](../experiments/ctdet_coco_dla_2x.sh) | 8 | 92 | 19 / 36 / 248 | 37.4 / 39.2 / 41.7 | [model](https://drive.google.com/open?id=1pl_-ael8wERdUREEnaIfqOV_VF2bEVRT) |
|[ctdet\_coco\_resdcn101](../experiments/ctdet_coco_resdcn101.sh)| 8 | 65 | 22 / 40 / 259 | 34.6 / 36.2 / 39.3 | [model](https://drive.google.com/open?id=1bTJCbAc1szA9lWU-fvVw52lqR3U2TTry) |
|[ctdet\_coco\_resdcn18](../experiments/ctdet_coco_resdcn18.sh) | 4 | 28 | 7 / 14 / 81 | 28.1 / 30.0 / 33.2 | [model](https://drive.google.com/open?id=1b-_sjq1Pe_dVxt5SeFmoadMfiPTPZqpz) |
|[exdet\_coco\_hg](../experiments/exdet_coco_hg.sh) | 5 |215 | 134 / 246/1340 | 35.8 / 39.8 / 42.4 | [model](https://drive.google.com/open?id=1-5bT5ZF8bXriJ-wAvOjJFrBLvZV2-mlV) |
|[exdet\_coco\_dla](../experiments/exdet_coco_dla.sh) | 8 |133 | 51 / 90 / 481 | 33.0 / 36.5 / 38.5 | [model](https://drive.google.com/open?id=1PFcEqN0KjFuq9XaqzB7TkVD3pvXZx04e) |
#### Notes
- All models are trained on COCO train 2017 and evaluated on val 2017.
- We show test time and AP with no augmentation / flip augmentation / multi scale (0.5, 0.75, 1, 1.25, 1.5) augmentation.
- Results on COCO test-dev can be found in the paper or add `--trainval` for `test.py`.
- exdet is our re-implementation of [ExtremeNet](https://github.com/xingyizhou/ExtremeNet). The testing does not include edge aggregation.
- For dla and resnets, `1x` means the training schedule that train 140 epochs with learning rate dropped 10 times at the 90 and 120 epoch (following [SimpleBaseline](https://github.com/Microsoft/human-pose-estimation.pytorch)). `2x` means train 230 epochs with learning rate dropped 10 times at the 180 and 210 epoch. The training schedules are **not** carefully investigated.
- The hourglass trained schedule follows [ExtremeNet](https://github.com/xingyizhou/ExtremeNet): trains 50 epochs (approximately 250000 iterations in batch size 24) and drops learning rate at the 40 epoch.
- Testing time include network forwarding time, decoding time, and nms time (for ExtremeNet).
- We observed up to 0.4 AP performance jitter due to randomness in training.
### Pascal VOC
| Model |GPUs| Train time (h)| Test time (ms) | mAP | Download |
|---------------------------------|----|---------------|----------------|------|-----------|
|[ctdet\_pascal\_dla\_384](../experiments/ctdet_pascal_dla_384.sh) | 1 |15 | 20 | 79.3 | [model](https://drive.google.com/open?id=1IC3FZkxAQHm2rxoIGmS4YluYpZxwYkJf) |
|[ctdet\_pascal\_dla\_512](../experiments/ctdet_pascal_dla_512.sh) | 2 |15 | 30 | 80.7 | [model](https://drive.google.com/open?id=1jIfK9EyqzNcupxGsp3YRnEiewrIG4_Ma) |
|[ctdet\_pascal\_resdcn18\_384](../experiments/ctdet_pascal_resdcn18_384.sh) | 1 |3 | 7 | 72.6 | [model](https://drive.google.com/open?id=1Kq27D1uoPZK42j2alDWmCGyqRU2ob1BX) |
|[ctdet\_pascal\_resdcn18\_512](../experiments/ctdet_pascal_resdcn18_512.sh) | 1 |5 | 10 | 75.7 | [model](https://drive.google.com/open?id=1MRUJTTJ4-ZDN0Y-zQOqQBqjrQMcXFzet) |
|[ctdet\_pascal\_resdcn101\_384](../experiments/ctdet_pascal_resdcn101_384.sh)| 2 |7 | 22 | 77.1 | [model](https://drive.google.com/open?id=11YXE04zILuXA5-kaYQEEg0ljNKBe6GPO) |
|[ctdet\_pascal\_resdcn101\_512](../experiments/ctdet_pascal_resdcn101_512.sh)| 4 |7 | 33 | 78.7 | [model](https://drive.google.com/open?id=1xhEf-a_y2Di6YdyPpCIj0-kVFjQvDf9N) |
#### Notes
- All models are trained on trainval 07+12 and tested on test 2007.
- Flip test is used by default.
- Training schedule: train for 70 epochs with learning rate dropped 10 times at the 45 and 60 epoch.
- We observed up to 1 mAP performance jitter due to randomness in training.
## Human pose estimation
### COCO
| Model | GPUs |Train time(h)| Test time (ms) | AP | Download |
|--------------------------|------|-------------|----------------|-------------|-----------|
|[multi\_pose\_hg_1x](../experiments/multi_pose_hg_1x.sh) | 5 |62 | 151 | 58.7 | [model](https://drive.google.com/open?id=1HBB5KRaSj-m-vtpGESm7_3evNP5Y84RS) |
|[multi\_pose\_hg_3x](../experiments/multi_pose_hg_3x.sh) | 5 |188 | 151 | 64.0 | [model](https://drive.google.com/open?id=1n6EvwhTbz7LglVXXlL9irJia7YuakHdB) |
|[multi\_pose\_dla_1x](../experiments/multi_pose_dla_1x.sh) | 8 |30 | 44 | 54.7 | [model](https://drive.google.com/open?id=1VeiRtuXfCbmhQNGV-XWL6elUzpuWN-4K) |
|[multi\_pose\_dla_3x](../experiments/multi_pose_dla_3x.sh) | 8 |70 | 44 | 58.9 | [model](https://drive.google.com/open?id=1PO1Ax_GDtjiemEmDVD7oPWwqQkUu28PI) |
#### Notes
- All models are trained on keypoint train 2017 images which contains at least one human with keypoint annotations (64115 images).
- The evaluation is done on COCO keypoint val 2017 (5000 images).
- Flip test is used by default.
- The models are fine-tuned from the corresponding center point detection models.
- Dla training schedule: `1x`: train for 140 epochs with learning rate dropped 10 times at the 90 and 120 epoch.`3x`: train for 320 epochs with learning rate dropped 10 times at the 270 and 300 epoch.
- Hourglass training schedule: `1x`: train for 50 epochs with learning rate dropped 10 times at the 40 epoch.`3x`: train for 150 epochs with learning rate dropped 10 times at the 130 epoch.
## 3D bounding box detection
#### Notes
- The 3dop split is from [3DOP](https://papers.nips.cc/paper/5644-3d-object-proposals-for-accurate-object-class-detection) and the suborn split is from [SubCNN](https://github.com/tanshen/SubCNN).
- No augmentation is used in testing.
- The models are trained for 70 epochs with learning rate dropped at the 45 and 60 epoch.
### KITTI 3DOP split
|Model |GPUs|Train time|Test time|AP-E|AP-M|AP-H|AOS-E|AOS-M|AOS-H|BEV-E|BEV-M|BEV-H| Download |
|------------|----|----------|---------|----|----|----|-----|-----|-----|-----|-----|-----|----------|
|[ddd_3dop](../experiments/ddd_3dop.sh)|2 | 7h | 31ms |96.9|87.8|79.2|93.9 |84.3 |75.7 |34.0 |30.5 |26.8 | [model](https://drive.google.com/open?id=1znsM6E-aVTkATreDuUVxoU0ajL1az8rz)|
### KITTI SubCNN split
|Model |GPUs|Train time|Test time|AP-E|AP-M|AP-H|AOS-E|AOS-M|AOS-H|BEV-E|BEV-M|BEV-H| Download |
|------------|----|----------|---------|----|----|----|-----|-----|-----|-----|-----|-----|----------|
|[ddd_sub](../experiments/ddd_sub.sh) |2 | 7h | 31ms |89.6|79.8|70.3|85.7 |75.2 |65.9 |34.9 |27.7 |26.4 | [model](https://drive.google.com/open?id=15XuJxTxCBnA8O37M_ghjppnWmVnjC0Hp)|