Skip to content

Latest commit

 

History

History
80 lines (61 loc) · 3.43 KB

TODO.md

File metadata and controls

80 lines (61 loc) · 3.43 KB
Raw

TODO

Agregar ejemplos de torchvision para lee una imagen y marcar los bounding boxes. Hacer un ejemplo que lea la imagen, la transforme en tensor y la pase al dibujo. Poner cabeceras a todos los archivos Completar el ciclo de entrenamiento con el archivo .pth Pasar al nuevo Github con fotos.

Ejemplos de conversiones entre vectores

Resolver el warning C:\Users\Usuari\development_env\object_detector_venv\lib\site-packages\torchvision\models_utils.py:223: UserWarning: Arguments other than a weight enum or None for 'weights' are deprecated since 0.13 https://www.geeksforgeeks.org/how-to-read-a-jpeg-or-png-image-in-pytorch/

Common operations made with Tensors

https://github.com/Tgaaly/pytorch-cheatsheet https://kunpengsong.github.io/2019/06/Pytorch-cheat-sheet/ https://stackoverflow.com/questions/43327668/looping-over-a-tensor

PENDING TASKS

  1. output with merged masks in binary [OK] test_mask_02.py
  2. output with colour image and binary mask overlapped [OK] test_mask_02.py
  3. output with bounding box + mask detected + coloured image [OK] test_mask_03.py
  4. process results from model with bounding box + mask detected + coloured image mask + binary mask. [OK] test_mask_04.py
  5. Process results from model and merge the mask
  6. Process all the images from a folder and merge the mask.
  7. bounding boxes
  8. Filter detections by threshold and label
  9. Create bounding box with label Filter by label name. Colour fixed, random colours Crete own dataset with folders Organise routines. Train with own labels, modify type of labels Test inference time Load saved and trained model. Save detected bounding boxes to files.

Training

Data augmentation

Dataset management

GPU settings in Windows 10

Use of GPU device in Windows 10. [OK]. Solved, it is needed to install CUDA 11.7 and then. Train in Linux Train in Windows. Create own dataset.

Tensor conversion

https://discuss.pytorch.org/t/how-to-use-pytorch-tensor-object-in-opencv-without-convert-to-numpy-array/66000/7 https://discuss.pytorch.org/t/pytorch-pil-to-tensor-and-vice-versa/6312 https://www.google.com/search?client=firefox-b-d&q=ndarray+to+tensor+pytorch https://pydlt.readthedocs.io/en/latest/source/quickstart/images.html https://www.tutorialspoint.com/how-to-convert-an-image-to-a-pytorch-tensor https://stackoverflow.com/questions/66237451/convert-numpy-ndarray-to-pil-and-and-convert-it-to-tensor https://www.geeksforgeeks.org/converting-an-image-to-a-torch-tensor-in-python/

Images are 1025 * 205 in .png format Add links in README.md to examples.

TRAIN MODELS LINKS

Apple Fruit Scab Recognition using Deep Learning and PyTorch https://hackernoon.com/pytorch-vs-tensorflow-who-has-more-pre-trained-deep-learning-models https://colab.research.google.com/github/amalphonse/SPAIC_sg_fruit_360/blob/master/Ivy_Fruits360_With_Pytorch.ipynb#scrollTo=1B2FZeRqfiYA

Installing libraries

https://pytorch.org/get-started/locally/ pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu116 pip3 install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu117 pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117

NVIDIA Deep Learning Examples

https://github.com/NVIDIA/DeepLearningExamples/tree/master/PyTorch/Segmentation/MaskRCNN

https://catalog.ngc.nvidia.com/orgs/nvidia/collections/nvidia_dlfw