Skip to content
This repository has been archived by the owner on Jul 2, 2021. It is now read-only.

Update readme #872

Merged
merged 7 commits into from
May 17, 2019
Merged

Update readme #872

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
fcbcf7d
update detection/README.md
knorth55 May 15, 2019
5706a33
update instance_segmentation/README.md
knorth55 May 15, 2019
70ea040
update semantic_segmentation readme
knorth55 May 16, 2019
ed24ab0
refactor instance_segmentation README
knorth55 May 15, 2019
af1d091
fix typo in README.md
knorth55 May 15, 2019
de755b2
update instance_segmentation/README.md
knorth55 May 15, 2019
dcc5d4a
fix typo in instance_segmentation/README.md
knorth55 May 16, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 5 additions & 0 deletions examples/detection/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,6 +29,9 @@ For the details, please check the documents and examples of each model.
You can reproduce these scores by the following command.
```
$ python eval_detection.py --dataset voc [--model faster_rcnn|ssd300|ssd512|yolo_v2|yolo_v2_tiny|yolo_v3] [--pretrained-model <model_path>] [--batchsize <batchsize>] [--gpu <gpu>]
# with multiple GPUs
$ mpiexec -n <#gpu> python eval_detection_multi.py --dataset voc [--model faster_rcnn|ssd300|ssd512|yolo_v2|yolo_v2_tiny|yolo_v3] [--pretrained-model <model_path>] [--batchsize <batchsize>]

```

### MS COCO2017 Val
Expand All @@ -41,6 +44,8 @@ $ python eval_detection.py --dataset voc [--model faster_rcnn|ssd300|ssd512|yolo
You can reproduce these scores by the following command.
```
$ python eval_detection.py --dataset coco [--model faster_rcnn_fpn_resnet50|faster_rcnn_fpn_resnet101] [--pretrained-model <model_path>] [--batchsize <batchsize>] [--gpu <gpu>]
# with multiple GPUs
$ mpiexec -n <#gpu> python eval_detection_multi.py --dataset coco [--model faster_rcnn_fpn_resnet50|faster_rcnn_fpn_resnet101] [--pretrained-model <model_path>] [--batchsize <batchsize>]
```

## Visualization of models
Expand Down
16 changes: 10 additions & 6 deletions examples/instance_segmentation/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,19 +10,21 @@ For the details, please check the documents and examples of each model.

## Performance

### SBD Test
### SBD Val

| Model | FPS | mAP@0.5 | mAP@0.7 |
|:-:|:-:|:-:|:-:|
| FCIS ResNet101 | | 64.1 % | 51.2 % |

You can reproduce these scores by the following command.

```bash
$ python eval_sbd.py [--model fcis_resnet101] [--pretrained-model <model_path>] [--gpu <gpu>]
```
$ python eval_instance_segmentation.py --dataset sbd [--model fcis_resnet101] [--pretrained-model <model_path>] [--batchsize <batchsize>] [--gpu <gpu>]
# with multiple GPUs
$ mpiexec -n <#gpu> python eval_instance_segmentation_multi.py --dataset sbd [--model fcis_resnet101] [--pretrained-model <model_path>] [--batchsize <batchsize>]
```

### COCO Test
### MS COCO2017 Val

| Model | FPS | mAP/iou@[0.5:0.95] | mAP/iou@[0.5:0.95] \(small) | mAP/iou@[0.5:0.95] \(medium) | mAP/iou@[0.5:0.95] \(large) |
|:-:|:-:|:-:|:-:|:-:|:-:|
Expand All @@ -32,8 +34,10 @@ $ python eval_sbd.py [--model fcis_resnet101] [--pretrained-model <model_path>]

You can reproduce these scores by the following command.

```bash
$ python eval_coco.py [--model fcis_resnet101|mask_rcnn_fpn_resnet50|mask_rcnn_fpn_resnet101] [--pretrained-model <model_path>] [--gpu <gpu>]
```
$ python eval_instance_segmentation.py --dataset coco [--model fcis_resnet101|mask_rcnn_fpn_resnet50|mask_rcnn_fpn_resnet101] [--pretrained-model <model_path>] [--batchsize <batchsize>] [--gpu <gpu>]
# with multiple GPUs
$ mpiexec -n <#gpu> python eval_instance_segmentation_multi.py --dataset coco [--model fcis_resnet101|mask_rcnn_fpn_resnet50|mask_rcnn_fpn_resnet101] [--pretrained-model <model_path>] [--batchsize <batchsize>]
```

## Notes on writing your own evaluation code
Expand Down
39 changes: 31 additions & 8 deletions examples/semantic_segmentation/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,36 +17,58 @@ The scores are mIoU.
|:-:|:-:|:-:|:-:|
| PSPNet w/ Dilated ResNet50 | fine only (3K) | 76.9 % [2] | 73.99 % |
| PSPNet w/ Dilated ResNet101 | fine only (3K) | 77.9 % [2] | 76.01 % |
| DeepLab V3+ w/ Xception65 | fine only (3K) | 79.12 % [4] | 79.14 % |


Example
You can reproduce these scores by the following command.

```
$ python eval_semantic_segmentation.py --gpu <GPU> --dataset cityscapes --model pspnet_resnet101
$ python eval_semantic_segmentation.py --dataset cityscapes [--model pspnet_resnet50|pspnet_resnet101|deeplab_v3plus_xception65] [--pretrained-model <model_path>] [--batchsize <batchsize>] [--gpu <gpu>]
# with multiple GPUs
$ mpiexec -n <#gpu> python eval_semantic_segmentation_multi.py --dataset cityscapes --model pspnet_resnet101
$ mpiexec -n <#gpu> python eval_semantic_segmentation_multi.py --dataset cityscapes [--model pspnet_resnet50|pspnet_resnet101|deeplab_v3plus_xception65] [--pretrained-model <model_path>] [--batchsize <batchsize>]
```

### ADE20k

| Base model | Reference | ChainerCV |
|:-:|:-:|:-:|
| Dilated ResNet50 | 41.68 % [1] | 34.97 % |
| Dilated ResNet101 | | 36.55 % |
| PSPNet w/ Dilated ResNet50 | 41.68 % [1] | 34.97 % |
| PSPNet w/ Dilated ResNet101 | | 36.55 % |
| DeepLab V3+ w/ Xception65 | | 42.52 % |

You can reproduce these scores by the following command.

```
$ python eval_semantic_segmentation.py --gpu <GPU> --dataset ade20k --model pspnet_resnet101
$ python eval_semantic_segmentation.py --dataset ade20k [--model pspnet_resnet50|pspnet_resnet101|deeplab_v3plus_xception65] [--pretrained-model <model_path>] [--batchsize <batchsize>] [--gpu <gpu>]
# with multiple GPUs
$ mpiexec -n <#gpu> python eval_semantic_segmentation_multi.py --dataset ade20k [--model pspnet_resnet50|pspnet_resnet101|deeplab_v3plus_xception65] [--pretrained-model <model_path>] [--batchsize <batchsize>]
```

### VOC2012 val

| Base model | Reference | ChainerCV |
|:-:|:-:|:-:|
| DeepLab V3+ w/ Xception65 | 82.36 % [4] | 82.36 % |

You can reproduce these scores by the following command.

```
$ python eval_semantic_segmentation.py --dataset voc [--model deeplab_v3plus_xception65] [--pretrained-model <model_path>] [--batchsize <batchsize>] [--gpu <gpu>]
# with multiple GPUs
$ mpiexec -n <#gpu> python eval_semantic_segmentation_multi.py --dataset voc [--model deeplab_v3plus_xception65] [--pretrained-model <model_path>] [--batchsize <batchsize>]
```

### CamVid

| Model | Reference | ChainerCV |
|:-:|:-:|:-:|
| SegNet | 46.3 % [3] | 49.4 % |

You can reproduce these scores by the following command.

```
$ python eval_semantic_segmentation.py --gpu <GPU> --dataset camvid --model segnet
$ python eval_semantic_segmentation.py --dataset camvid [--model segnet] [--pretrained-model <model_path>] [--batchsize <batchsize>] [--gpu <gpu>]
# with multiple GPUs
$ mpiexec -n <#gpu> python eval_semantic_segmentation_multi.py --dataset camvid [--model segnet] [--pretrained-model <model_path>] [--batchsize <batchsize>]
```


Expand All @@ -55,3 +77,4 @@ $ python eval_semantic_segmentation.py --gpu <GPU> --dataset camvid --model segn
1. Hengshuang Zhao et al. "Pyramid Scene Parsing Network" CVPR 2017.
2. https://github.com/holyseven/PSPNet-TF-Reproduce (Validation scores for Cityscapes are lacking in the original paper)
3. Vijay Badrinarayanan et al. "SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation." PAMI, 2017.
4. Liang-Chieh Chen et al. "Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation" ECCV, 2018.