# MMAction2
https://github.com/open-mmlab/mmaction2

https://github.com/open-mmlab/mmaction2/blob/main/demo/README.md

https://mmaction2.readthedocs.io/zh-cn/latest/get_started/installation.html

https://congee524-mmaction2.readthedocs.io/zh-cn/latest/demo.html

---

### DEMO測試
1. conda activate openmmlab
2. cd D:\mmaction2
3. python demo/demo_inferencer.py demo/zelda.mp4 --rec tsn --print-result --vid-out-dir zelda_out --label-file tools/data/kinetics/label_map_k400.txt

### 數據集瀏覽
* python tools/visualizations/browse_dataset.py configs/recognition/tsn/tsn_imagenet-pretrained-r50_8xb32-1x1x3-100e_kinetics400-rgb_copy.py --output-dir browse_out --mode pipeline

### 訓練
* python tools/train.py configs/recognition/tsn/tsn_imagenet-pretrained-r50_8xb32-1x1x3-100e_kinetics400-rgb_copy.py
->調整訓練次數(epoch)

### 驗證
* python tools/test.py configs/recognition/tsn/tsn_imagenet-pretrained-r50_8xb32-1x1x3-100e_kinetics400-rgb_copy.py work_dirs/tsn_imagenet-pretrained-r50_8xb32-1x1x3-100e_kinetics400-rgb_copy/best_acc_top1_epoch_1.pth

### 數據集位置
* D:\mmaction2\data\finger(單一手勢)
* D:\mmaction2\data\action(數據集改連貫性動作)

---

### 準備數據集
>數據集需放在MMACTION2/data下
>(train/val)

>將數據集轉換為現有的數據集格式(txt)，有三種
* **幀標註（rawframe annotation）**
->分別是rawframe directory of relative path/total frames/label
dataset_type = 'RawframeDataset'
* **視訊標註（video annotation）**
->分別是filepath of relative path/label
dataset_type = 'VideoDataset'
* **ActivityNet 標註**
->標註檔是一個json檔，每個鍵是一個視頻名

---

### 模型推理(+camera)
* python demo/demo_test.py D:/mmaction2/camera/recorded_video.avi --rec work_dirs/tsn_imagenet-pretrained-r50_8xb32-1x1x3-100e_kinetics400-rgb_copy/tsn_imagenet-pretrained-r50_8xb32-1x1x3-100e_kinetics400-rgb_copy.py --print-result --vid-out-dir zelda_out

---

### 實時動作識別
**1. DEMO**
* python demo/webcam_demo.py configs/recognition/tsn/tsn_imagenet-pretrained-r50_8xb32-1x1x3-100e_kinetics400-rgb.py tsn_imagenet-pretrained-r50_8xb32-1x1x8-100e_kinetics400-rgb_20220906-2692d16c.pth tools/data/kinetics/label_map_k400.txt --average-size 1 --threshold 0.8

**2. Train**
* python demo/webcam_demo.py configs/recognition/tsn/tsn_imagenet-pretrained-r50_8xb32-1x1x3-100e_kinetics400-rgb_copy.py work_dirs/tsn_imagenet-pretrained-r50_8xb32-1x1x3-100e_kinetics400-rgb_copy/epoch_50.pth finger.txt --average-size 5 --threshold 0.5

* best：python demo/webcam_demo.py configs/recognition/tsn/tsn_imagenet-pretrained-r50_8xb32-1x1x3-100e_kinetics400-rgb_copy.py work_dirs/tsn_imagenet-pretrained-r50_8xb32-1x1x3-100e_kinetics400-rgb_copy/finger_1/epoch_50.pth finger.txt --average-size 5 --threshold 0.5

**3. Spatio-Temporal Action**
* python demo/webcam_demo_spatiotemporal_det.py --input-video 0 --config configs/detection/slowonly/slowonly_kinetics400-pretrained-r101_8xb16-8x8x1-20e_ava21-rgb.py --checkpoint https://download.openmmlab.com/mmaction/detection/ava/slowonly_omnisource_pretrained_r101_8x8x1_20e_ava_rgb/slowonly_omnisource_pretrained_r101_8x8x1_20e_ava_rgb_20201217-16378594.pth --det-config demo/demo_configs/faster-rcnn_r50_fpn_2x_coco_infer.py --det-checkpoint http://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_2x_coco/faster_rcnn_r50_fpn_2x_coco_bbox_mAP-0.384_20200504_210434-a5d8aa15.pth --det-score-thr 0.9 --action-score-thr 0.5 --label-map tools/data/ava/label_map.txt --predict-stepsize 40 --output-fps 65 --show