paddleX训练好的模型,不能直接用到PaddleDetection吗

PaddlePaddle / PaddleDetection

Object Detection toolkit based on PaddlePaddle. It supports object detection, instance segmentation, multiple object tracking and real-time multi-person keypoint detection.

Apache License 2.0

12.86k stars 2.9k forks source link

paddleX训练好的模型,不能直接用到PaddleDetection吗 #2578

Open monkeycc opened 3 years ago

monkeycc commented 3 years ago

模型是 FasterRCNN

paddleX训练好的模型 model.yml

改名为 infer_cfg.yml

让PaddleDetection进行预测 deploy\python\infer.py

SUPPORT_MODELS 中没发现有FasterRCNN 要选择RCNN？？

SUPPORT_MODELS = { 'YOLO', 'SSD', 'RetinaNet', 'EfficientDet', 'RCNN', 'Face', 'TTF', 'FCOS', 'SOLOv2', }

FasterRCNN中，需要将 with_background=true 且 num_classes+1 已经更改

对 infer_cfg.yml 进行增加更改还是各种错误

是我有些东西没改对还是需要在PaddleDetection中重新训练

liuhuiCNN commented 3 years ago

（1）SUPPORT_MODELS中RCNN是指FasterRCNN系列模型；（2）paddleX是在PaddleDetection上的封装，应该是不需要重新训练。（3）此外请注意PaddleX和PaddleDetection对应版本关系。

monkeycc commented 3 years ago

Model: FasterRCNN Transforms:

Normalize: mean:
- 0.485
- 0.456
- 0.406 std:
- 0.229
- 0.224
- 0.225
ResizeByShort: max_size: 1333 short_size: 800
Padding: coarsest_stride: 32 target_size: null TransformsMode: RGB _Attributes: eval_metrics: bbox_map: 72.29197703403773 fixed_input_shape: null labels:
- background
- xx1
- xx2
- xx3 model_type: detector num_classes: 4 _ModelInputsOutputs: test_inputs:
- - image
- image
- - im_info
- im_info
- - im_shape
- im_shape test_outputs:
- - bbox
- multiclass_nms_0.tmp_0 _init_params: anchor_sizes:
- 32
- 64
- 128
- 256
- 512 aspect_ratios:
- 0.5
- 1.0
- 2.0 backbone: ResNet50 num_classes: 14 with_fpn: true completed_epochs: 0 status: Infer version: 1.2.8

这个是paddleX训练好的模型 model.yml 现在要怎么转换为PaddleDetection的deploy\python\infer.py 能预测的 infer_cfg.yml 弄了一天没弄出来谢谢

zhengwangwang commented 1 year ago

没人了吗?我这个问题找了几天了,还是没解决

monkeycc commented 1 year ago

@zhengwangwang paddleX 多久没更新了

据说新版本会支持FastDeploy