[MD] Transforms Unification Feature - Updates models to use unified dataset transforms · Pull Request !2782 · MindSpore/models

**What type of PR is this?**
/kind task

**What does this PR do / why do we need it**:
MindData dataset has provided the Transforms Unification feature to replace legacy c_transforms.py and py_transforms.py with unified transforms.py .
Need to update models to use the unified transforms.

Fixes # https://e.gitee.com/mind_spore/dashboard?issue=I52FD5

**Special notes for your reviewers**:

1) Updates done to change from old legacy transforms to new unified transforms:

a) Access/import for unified transforms:
Use new and recommended format:
```
import mindspore.dataset.transforms as transforms
import mindspore.dataset.vision as vision
```
Old legacy and deprecated format examples:
```
import mindspore.dataset.transforms.c_transforms as c_trans
import mindspore.dataset.transforms.py_transforms as py_trans
import mindspore.dataset.vision.c_transforms as c_vision
import mindspore.dataset.vision.py_transforms as py_vision
```

b) Normalize op
    - for legacy Python Normalize op usage, replace with unified Normalize op usage and set new arg `is_hwc=False`
    - note: for legacy C++ Normalize op usage, when replacing with unified Normalize op usage, use default `is_hwc=True` or set explicitly (for added clarity in model scripts)

c) Decode op
    - for legacy Python Decode op usage, replace with unified Decode op usage and set new arg `is_pil=True`
    - note: for legacy C++ Decode op usage, use default `is_pil=False` or set explicitly (for added clarity in model scripts)

d) ToType op
    - for legacy vision ToType op usage, replace with unified data transform TypeCast op usage
    - 1 usage in research/cv/FaceRecognition/src/dataset_factory.py

e) Other dataset ops - no impact in models:
No usage of these legacy Python ops in models:
    - Cutout op
    - NormalizePad (fyi - 1 C++ op usage in official/cv/resnet/gpu_resnet_benchmark.py)
    - OneHotOp 
    - RgbToBgr op

2) Number of files in the PR
    - benchmark = 3
    - official = 93
    - research = 196
    - utils = 1
    - jenkins filter = 1
    - TOTAL = 294

3) For candidate test coverage:
a) official/cv files with unified Normalize `is_hwc=False` op:
```
./official/cv/fastscnn/eval.py
./official/cv/fastscnn/modelarts/start_train.py
./official/cv/fastscnn/train.py
./official/cv/east/detect.py
./official/cv/patchcore/src/dataset.py
./official/cv/patchcore/preprocess.py
./official/cv/efficientnet/src/transform.py
./official/cv/pvnet/eval.py
./official/cv/pvnet/src/dataset.py
./official/cv/psenet/src/dataset.py
```
- note: These are just the updated official/cv files; there are many updated research/cv files.
b) cv files with unified Decode `is_pil=True` op:
```
./official/cv/yolov5/src/transforms.py
./research/cv/ManiDP/src/dataset.py
./research/cv/ManiDP/src/dataset.py
./research/cv/wave_mlp/src/dataset.py
./research/cv/HireMLP/src/dataset.py
./research/cv/resnet50_adv_pruning/src/pet_dataset.py
./research/cv/DRNet/src/dataset.py
./research/cv/ghostnet_quant/src/dataset.py
./research/cv/tinynet/src/dataset.py
./research/cv/tinynet/src/dataset.py
./research/cv/FDA-BNN/src/dataset.py
./research/cv/FDA-BNN/src/dataset.py
./research/cv/FaceAttribute/src/dataset_eval.py
./research/cv/FaceAttribute/src/dataset_train.py
./research/cv/FaceAttribute/preprocess.py
./research/cv/yolov3_tiny/src/transforms.py
./research/cv/FaceDetection/src/data_preprocess.py
./research/cv/FaceDetection/src/data_preprocess.py
./research/cv/FaceDetection/preprocess.py
./research/cv/HourNAS/src/dataset.py
./research/cv/HourNAS/src/dataset.py
./research/cv/renas/src/dataset.py
./research/cv/renas/src/dataset.py
./research/cv/CMT/src/dataset.py
./research/cv/AutoSlim/src/dataset.py
```

4) FYI - some of the updates were done semi-automatically using script commands.
Sample update commands used (for official/cv):
```
perl -pi -e 's/mindspore.dataset.transforms.c_transforms/mindspore.dataset.transforms/g' official/*/*/*.py
perl -pi -e 's/mindspore.dataset.transforms.c_transforms/mindspore.dataset.transforms/g' official/*/*/*/*.py
perl -pi -e 's/mindspore.dataset.transforms.c_transforms/mindspore.dataset.transforms/g' official/*/*/*/*/*.py

perl -pi -e 's/ds.transforms.c_transforms/ds.transforms.transforms/g' official/*/*/*.py
perl -pi -e 's/ds.transforms.c_transforms/ds.transforms.transforms/g' official/*/*/*/*.py
perl -pi -e 's/ds.transforms.c_transforms/ds.transforms.transforms/g' official/*/*/*/*/*.py
```

**CheckList**:

- [x] I have added correct copyrights for every code file.
- [x] I have removed all the redundant code and comments.
- [x] I have updated or added the `requirements.txt` for the third-party libraries you need.
- [x] I have made sure that I won't expose any personal information such as local path with user name, local IP, etc.
- [x] I have commented my code, particularly in hard-to-understand areas. All the comments in code files are in English.
- [x] I have made corresponding changes to the documentation.
- [x] I have squashed all the commits into one.
- [ ] I have test and ascertained the effect of my change in all related cases.
    - [ ] Different hardware: `CPU`, `GPU`, `Ascend910`, `Ascend310`, `Ascend701`.
    - [ ] Different mode: `GRAPH_MODE`, `PYNATIVE_MODE`.
    - [ ] Different system: `Linux`, `Windows`, `MAC`.
    - [ ] Different number of cluster: `1pc`, `8pcs`.
- [x] I have checked the following usual mistakes:
    - [x] It is recommended to use `bash` instead of `sh`.
    - [x] Don't use `type=bool` in argparse.
- [ ] If you are contributing a new model, please check:
    - [ ] I have added at least one `README`
    - [ ] The models implemented for the cloud could work offline on the local server as well.
    - [ ] I have made sure the changes for new device target will not make differences on the original one.
- [ ] If you are contributing a new function, please check:
    - [ ] I have added the corresponding information in README.
    - [ ] I have notify the maintenance team about this function.

MindSpore/models .gitee-modal { width: 500px !important; }

搜索帮助

MindSpore/models