- [Description of Random Situation](#description-of-random-situation)
- [ModelZoo Homepage](#modelzoo-homepage)
# [MaskRCNN Description](#contents)
MaskRCNN is a conceptually simple, flexible, and general framework for object instance segmentation. The approach efficiently detects objects in an image while simultaneously generating a high-quality segmentation mask for each instance. The method, called Mask R-CNN, extends Faster R-CNN by adding a branch for predicting an object mask in
parallel with the existing branch for bounding box recognition. Mask R-CNN is simple to train and adds only a small overhead to Faster R-CNN, running at 5 fps. Moreover, Mask R-CNN is easy to generalize to other tasks, e.g., allowing to estimate human poses in the same framework.
It shows top results in all three tracks of the COCO suite of challenges, including instance segmentation, boundingbox object detection, and person keypoint detection. Without bells and whistles, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO 2016 challenge winners.
# [Model Architecture](#contents)
MaskRCNN is a two-stage target detection network. It extends FasterRCNN by adding a branch for predicting an object mask in parallel with the existing branch for bounding box recognition.This network uses a region proposal network (RPN), which can share the convolution features of the whole image with the detection network, so that the calculation of region proposal is almost cost free. The whole network further combines RPN and mask branch into a network by sharing the convolution features.
This network uses MobileNetV1 as the backbone of the MaskRCNN network.
[Paper](http://cn.arxiv.org/pdf/1703.06870v3): Kaiming He, Georgia Gkioxari, Piotr Dollar and Ross Girshick. "MaskRCNN"
# [Dataset](#contents)
Note that you can run the scripts based on the dataset mentioned in original paper or widely used in relevant domain/network architecture. In the following sections, we will introduce how to run the scripts using the related dataset below.
- [COCO2017](https://cocodataset.org/) is a popular dataset with bounding-box and pixel-level stuff annotations. These annotations can be used for scene understanding tasks like semantic segmentation, object detection and image captioning. There are 118K/5K images for train/val.
- Dataset size: 19G
- Train: 18G, 118000 images
- Val: 1G, 5000 images
- Annotations: 241M, instances, captions, person_keypoints, etc.
Each row is an image annotation split by spaces. The first column is a relative path of image, followed by columns containing box and class information in the format [xmin,ymin,xmax,ymax,class]. We read image from an image path joined by the `IMAGE_DIR`(dataset directory) and the relative path in `ANNO_PATH`(the TXT file path), which can be set in `config.py`.
3. Execute train script.
After dataset preparation, you can start training as follows:
```
# distributed training
sh run_distribute_train.sh [RANK_TABLE_FILE] [PRETRAINED_CKPT]
# standalone training
sh run_standalone_train.sh [PRETRAINED_CKPT]
```
Note:
1. To speed up data preprocessing, MindSpore provide a data format named MindRecord, hence the first step is to generate MindRecord files based on COCO2017 dataset before training. The process of converting raw COCO2017 dataset to MindRecord format may take about 4 hours.
2. For distributed training, a [hccl configuration file](https://gitee.com/mindspore/mindspore/tree/master/model_zoo/utils/hccl_tools) with JSON format needs to be created in advance.
3. For large models like MaskRCNN, it's better to export an external environment variable `export HCCL_CONNECT_TIMEOUT=600` to extend hccl connection checking time from the default 120 seconds to 600 seconds. Otherwise, the connection could be timeout since compiling time increases with the growth of model size.
4. Execute eval script.
After training, you can start evaluation as follows:
```bash
# Evaluation
sh run_eval.sh [VALIDATION_JSON_FILE] [CHECKPOINT_PATH]
```
Note:
1. VALIDATION_JSON_FILE is a label json file for evaluation.
# [Script Description](#contents)
## [Script and Sample Code](#contents)
```shell
.
└─MaskRcnn
├─README.md # README
├─scripts # shell script
├─run_standalone_train.sh # training in standalone mode(1pcs)
├─run_distribute_train.sh # training in parallel mode(8 pcs)
└─run_eval.sh # evaluation
├─src
├─maskrcnn_mobilenetv1
├─__init__.py
├─anchor_generator.py # generate base bounding box anchors
├─bbox_assign_sample.py # filter positive and negative bbox for the first stage learning
├─bbox_assign_sample_stage2.py # filter positive and negative bbox for the second stage learning
├─mask_rcnn_mobilenetv1.py # main network architecture of maskrcnn
├─fpn_neck.py # fpn network
├─proposal_generator.py # generate proposals based on feature map
- Set options in `config.py`, including loss_scale, learning rate and network hyperparameters. Click [here](https://www.mindspore.cn/tutorial/training/zh-CN/master/use/data_preparation.html) for more information about dataset.
### [Training](#content)
- Run `run_standalone_train.sh` for non-distributed training of MaskRCNN model.
```bash
# standalone training
sh run_standalone_train.sh [PRETRAINED_MODEL]
```
### [Distributed Training](#content)
- Run `run_distribute_train.sh` for distributed training of Mask model.
```bash
sh run_distribute_train.sh [RANK_TABLE_FILE] [PRETRAINED_MODEL]
```
> hccl.json which is specified by RANK_TABLE_FILE is needed when you are running a distribute task. You can generate it by using the [hccl_tools](https://gitee.com/mindspore/mindspore/tree/master/model_zoo/utils/hccl_tools).
> As for PRETRAINED_MODEL, if not set, the model will be trained from the very beginning. Ready-made pretrained_models are not available now. Stay tuned.
> This is processor cores binding operation regarding the `device_num` and total processor numbers. If you are not expect to do it, remove the operations `taskset` in `scripts/run_distribute_train.sh`
Training result will be stored in the example path, whose folder name begins with "train" or "train_parallel". You can find checkpoint file together with result like the following in loss_rankid.log.
