forked from WongKinYiu/yolov7
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
update ymir-exc-sdk from 1.3.0 to 2.1.0
- Loading branch information
Showing
8 changed files
with
240 additions
and
129 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1,112 +1,79 @@ | ||
| # ymir-yolov5 | ||
| # ymir-yolov7 镜像说明文档 | ||
|
|
||
| ``` | ||
| docker pull youdaoyzbx/ymir-executor:ymir1.1.0-yolov7-cu111-tmi | ||
| docker build -t ymir/executor:yolov7-cu111-tmi --build-arg YMIR=1.1.0 -f ymir/docker/cuda111.dockerfile . | ||
| ``` | ||
|
|
||
|
|
||
| ## change log | ||
| - add `ymir` folder | ||
| - modify `train.py` to write `monitor.txt` and `result.yaml` | ||
| - modify `utils/datasets.py` to support ymir dataset format | ||
| ## 代码仓库 | ||
|
|
||
| > 参考[WongKinYiu/yolov7](https://github.com/WongKinYiu/yolov7) | ||
| - [modelai/ymir-yolov7](https://github.com/modelai/ymir-yolov7) | ||
|
|
||
| ## support ymir dataset format | ||
| - use `convert_ymir_to_yolov5` to generate `data.yaml` | ||
| ## 镜像地址 | ||
|
|
||
| - modify `utils/datasets.py` load `train.tsv`, `val.tsv` | ||
| ``` | ||
| # support ymir index file `train.tsv` and `val.tsv` | ||
| # class LoadImagesAndLabels(Dataset): | ||
| f = [] # image files | ||
| img2label_map = dict() # map image files to label files | ||
| for p in path if isinstance(path, list) else [path]: | ||
| p = Path(p) # os-agnostic | ||
| if p.is_file(): # file | ||
| with open(p, 'r') as t: | ||
| t = t.read().strip().splitlines() | ||
| for x in t: | ||
| # x = f'{image_path}\t{label_path}\n' | ||
| image_path, label_path = x.split() | ||
| f.append(image_path) | ||
| img2label_map[image_path] = label_path | ||
| else: | ||
| raise Exception(f'{prefix}{p} is not valid ymir index file') | ||
| # get label file path from image file path | ||
| def img2label_paths(img_paths, img2label_map={}): | ||
| return [img2label_map[img] for img in img_paths] | ||
| self.label_files = img2label_paths(self.img_files, img2label_map) | ||
| # support ymir label file | ||
| # convert ymir (xyxy) to yolov5 bbox format normalized (xc,yc,w,h) | ||
| # def cache_labels() | ||
| l_ymir = np.array(l, dtype=np.float32) | ||
| l = l_ymir.copy() | ||
| width, height = imagesize.get(im_file) | ||
| l[:,1:5:2] /= width # normalize x1,x2 | ||
| l[:,2:5:2] /= height # normalize y1,y2 | ||
| l[:,1] = (l_ymir[:,1] + l_ymir[:,3])/2/width | ||
| l[:,2] = (l_ymir[:,2] + l_ymir[:,4])/2/height | ||
| l[:,3] = (l_ymir[:,3] - l_ymir[:,1])/width | ||
| l[:,4] = (l_ymir[:,4] - l_ymir[:,2])/height | ||
| ``` | ||
|
|
||
| ## write monitor and training results | ||
| youdaoyzbx/ymir-executor:ymir2.1.0-yolov7-cu111-tmi | ||
| modify `train.py` | ||
| - import functions | ||
| ``` | ||
| from ymir.ymir_yolov5 import get_merged_config, YmirStage, get_ymir_process | ||
| from ymir_exc import monitor | ||
| from ymir_exc import result_writer as rw | ||
| # 此版本加载预训练权重需要epoch>原有epoch，否则会出错. | ||
| youdaoyzbx/ymir-executor:ymir2.0.0-yolov7-cu111-tmi | ||
| ``` | ||
|
|
||
| - write tensorboard results | ||
| ## 性能表现 | ||
|
|
||
| > 数据参考[WongKinYiu/yolov7](https://github.com/WongKinYiu/yolov7) | ||
| | Model | Test Size | AP<sup>test</sup> | AP<sub>50</sub><sup>test</sup> | AP<sub>75</sub><sup>test</sup> | batch 1 fps | batch 32 average time | | ||
| | :-- | :-: | :-: | :-: | :-: | :-: | :-: | | ||
| | [**YOLOv7**](https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7.pt) | 640 | **51.4%** | **69.7%** | **55.9%** | 161 *fps* | 2.8 *ms* | | ||
| | [**YOLOv7-X**](https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7x.pt) | 640 | **53.1%** | **71.2%** | **57.8%** | 114 *fps* | 4.3 *ms* | | ||
| | | | | | | | | | ||
| | [**YOLOv7-W6**](https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7-w6.pt) | 1280 | **54.9%** | **72.6%** | **60.1%** | 84 *fps* | 7.6 *ms* | | ||
| | [**YOLOv7-E6**](https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7-e6.pt) | 1280 | **56.0%** | **73.5%** | **61.2%** | 56 *fps* | 12.3 *ms* | | ||
| | [**YOLOv7-D6**](https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7-d6.pt) | 1280 | **56.6%** | **74.0%** | **61.8%** | 44 *fps* | 15.0 *ms* | | ||
| | [**YOLOv7-E6E**](https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7-e6e.pt) | 1280 | **56.8%** | **74.4%** | **62.1%** | 36 *fps* | 18.7 *ms* | | ||
|
|
||
|
|
||
| ## 训练参数 | ||
|
|
||
| | 超参数 | 默认值 | 类型 | 说明 | 建议 | | ||
| | - | - | - | - | - | | ||
| | hyper-parameter | default value | type | note | advice | | ||
| | shm_size | 128G | 字符串| 受ymir后台处理，docker image 可用共享内存 | 建议大小：镜像占用GPU数 * 32G | | ||
| | export_format | ark:raw | 字符串| 受ymir后台处理，ymir数据集导出格式 | - | | ||
| | model | yolov5s | 字符串 | yolov5模型，可选yolov5n, yolov5s, yolov5m, yolov5l等 | 建议：速度快选yolov5n, 精度高选yolov5l, yolov5x, 平衡选yolov5s或yolov5m | | ||
| | batch_size_per_gpu | 16 | 整数 | 每张GPU一次处理的图片数量 | 建议大小：显存占用<50% 可增加2倍加快训练速度 | | ||
| | workers_per_gpu | 4 | 整数 | 每张GPU对应的数据读取进程数 | - | | ||
| | epochs | 100 | 整数 | 整个数据集的训练遍历次数 | 建议：必要时分析tensorboard确定是否有必要改变，一般采用默认值即可 | | ||
| | img_size | 640 | 整数 | 输入模型的图像分辨率 | - | | ||
| | args_options | '--exist-ok' | 字符串 | yolov5命令行参数 | 建议：专业用户可用yolov5所有命令行参数 | | ||
| | save_weight_file_num | 1 | 整数 | 保存最新模型的数量 | - | | ||
| | sync_bn | False | 布尔型 | 是否同步各gpu上的归一化层 | 建议：开启以提高训练稳定性及精度 | | ||
| | cfg_file | cfg/training/yolov7-tiny.yaml | 文件路径 | 模型文件路径, 对应 `--cfg` | 参考[cfg/training](https://github.com/modelai/ymir-yolov7/tree/ymir/cfg/training) | | ||
| | hyp_file | data/hyp.scratch.tiny.yaml | 文件路径 | 超参数文件路径，对应 `--hyp` | 参考[data](https://github.com/modelai/ymir-yolov7/tree/ymir/data) | | ||
| | cache_images | True | 布尔 | 是否缓存图像 | 设置为True可加快训练速度 | | ||
|
|
||
|
|
||
| ## 推理参数 | ||
|
|
||
| | 超参数 | 默认值 | 类型 | 说明 | 建议 | | ||
| | - | - | - | - | - | | ||
| | hyper-parameter | default value | type | note | advice | | ||
| | img_size | 640 | 整数 | 模型的输入图像大小 | 采用32的整数倍，224 = 32*7 以上大小 | | ||
| | conf_thres | 0.25 | 浮点数 | 置信度阈值 | 采用默认值 | | ||
| | iou_thres | 0.45 | 浮点数 | nms时的iou阈值 | 采用默认值 | | ||
|
|
||
| ## 挖掘参数 | ||
|
|
||
| | 超参数 | 默认值 | 类型 | 说明 | 建议 | | ||
| | - | - | - | - | - | | ||
| | hyper-parameter | default value | type | note | advice | | ||
| | shm_size | 128G | 字符串| 受ymir后台处理，docker image 可用共享内存 | 建议大小：镜像占用GPU数 * 32G | | ||
| | img_size | 640 | 整数 | 模型的输入图像大小 | 采用32的整数倍，224 = 32*7 以上大小 | | ||
| | conf_thres | 0.25 | 浮点数 | 置信度阈值 | 采用默认值 | | ||
| | iou_thres | 0.45 | 浮点数 | nms时的iou阈值 | 采用默认值 | | ||
|
|
||
| ## 引用 | ||
| ``` | ||
| # change tensorboard writer log_dir | ||
| ymir_cfg = get_merged_config() | ||
| tb_writer = SummaryWriter(ymir_cfg.ymir.output.tensorboard_dir) # Tensorboard | ||
| ``` | ||
|
|
||
| - monitor training process | ||
| @article{wang2022yolov7, | ||
| title={{YOLOv7}: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors}, | ||
| author={Wang, Chien-Yao and Bochkovskiy, Alexey and Liao, Hong-Yuan Mark}, | ||
| journal={arXiv preprint arXiv:2207.02696}, | ||
| year={2022} | ||
| } | ||
| ``` | ||
| # for each epoch | ||
| monitor_gap = max(1, (epochs - start_epoch)//1000) | ||
| if rank in [-1, 0] and epoch % monitor_gap == 0: | ||
| monitor.write_monitor_logger(percent=get_ymir_process(stage=YmirStage.TASK, p=(epoch-start_epoch)/(epochs-start_epoch))) | ||
| ``` | ||
|
|
||
| - write `result.yaml` to save model weights and map50 | ||
| - optional: modify `utils/metrics.py` fitness() to save best map50 | ||
| ``` | ||
| def fitness(x): | ||
| # Model fitness as a weighted combination of metrics | ||
| w = [0.0, 0.0, 1.0, 0.0] # weights for [P, R, mAP@0.5, mAP@0.5:0.95] | ||
| return (x[:, :4] * w).sum(1) | ||
| ``` | ||
| ``` | ||
| if (epoch + 1) % opt.save_period == 0: | ||
| epoch_weight_file = wdir / 'epoch_{:03d}.pt'.format(epoch) | ||
| torch.save(ckpt, epoch_weight_file) | ||
| write_ymir_training_result(ymir_cfg, map50=float(results[2]), id=str(epoch), files=[str(epoch_weight_file)]) | ||
| ``` | ||
|
|
||
| - modify `start.py` save other output files | ||
| ``` | ||
| # save other files in output directory | ||
| write_ymir_training_result(cfg, map50=0, id='last', files=[]) | ||
| # if task done, write 100% percent log | ||
| monitor.write_monitor_logger(percent=1.0) | ||
| ``` | ||
|
|
||
| ## infer and mining | ||
| - view `ymir/start.py` "_run_infer()" for infer | ||
| - view `ymir/start.py` "_run_mining()" for mining | ||
|
|
||
| ## TODO | ||
| - [ ] multi-gpu mining and infer | ||
| - [ ] batch mining and infer |
Oops, something went wrong.