https://github.com/thaoshibe/lets-grab-a-chair-
Let's grab a chair!
https://github.com/thaoshibe/lets-grab-a-chair-
Last synced: 4 days ago
JSON representation
Let's grab a chair!
- Host: GitHub
- URL: https://github.com/thaoshibe/lets-grab-a-chair-
- Owner: thaoshibe
- Created: 2020-09-30T09:57:21.000Z (over 4 years ago)
- Default Branch: master
- Last Pushed: 2020-10-08T03:47:13.000Z (over 4 years ago)
- Last Synced: 2025-03-14T17:44:40.272Z (3 months ago)
- Size: 1.82 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
### Let's Grab A Chair!
---
##### Table of Contents
1. [Quick Start! Fun inference in tik-tak](#Quick-Start)
1. [Evaluation Guidelines](#Evaluation-Guidelines)
1. [Training Guidelines](#Training-Guidelines)
---
### Quick Start
Quick inference trained model. Sample reference image folder are `./sample`. All images in this folder are **in the wild** images
In all subsection, I suppose that you're in project folder `/home/ubuntu/adv-cv-class`
`pip install -r requirements.txt`
##### MaskRCNN
```
cd maskrcnn
python3.7 detect_chair.py
```
Predicted images are saved in form *pr_{}.png*.
##### Semantic Segmentation
```
cd semantic_segmentation
python3.7 test.py --datapath /home/ubuntu/thao-adv-cv-class/sample --test
```
Predicted images are saved in `/resnet50_fpn/test/{}.png`
##### Yolov5
```
cd yolov5
python3.7 detect.py --weights runs/exp20/weights/best.pt --source ../sample
```
Predicted imgs will be saved in *inference/output/{}.jpg*
---
### Evaluation Guidelines
##### **ADE20K Preparation**: `python ade20k_convert_validation.py`
This code will convert raw validation set to Chairs Validation Set
Original ADE20K Validation should be in `/home/ubuntu/ADE20K_2016_07_26/images/validation/**/*`
New Chairs Validation set will be saved in `/home/ubuntu/ADE20K_new/val_imgs` and `/home/ubuntu/ADE20K_new/val_segs`
Number of imgs: 2000 imgs (& 2000 masks)/ Estimated time: 1mins40s
##### **MaskRCNN**:
`python3.7 valid.py`
##### **Semantic Segmentation**:
`python3.7 valid.py --test`
##### **Yolov5**:
- Create label as txt file for YOLO:
`python3.7 create_yolo_validation.py`
- Evaluation
`python3.7 valid.py --img 640 --batch 1 --data ./data/chairs.yaml --weights runs/exp20/weights/best.pt --task val --conf-thres 0.01`
#### Training Guidelines
##### MaskRCNN & Semantic Segmentation:
- Prepare datasset: in `./semantic_segmentation/create_data.ipynb` to create `/home/ubuntu/ADE20K_Chairs_Full/train_imgs` (4827 imgs with segmentation
- Randomly add more ~8k images to `/home/ubuntu/ADE20K_Chairs_Full/train_imgs`
- Run `./semantic_segmentation/add_empty_label.ipynb` to create empty label for images without chairs (return zeros_mask)
- Train: `python3.7 train.py` in each folder
1. Yolov5
- Yolo need special txt file to store label. Simply run notebook `yolov5/create_yolo.ipynb`, all labels for YOLO will created.
- Train: `python3.7 train.py --img 640 --batch 8 --epochs 100 --data ./data/chairs.yaml --cfg ./models/yolov5x_chair.yaml --weights yolov5x.pt`
##### References
My code are heavily borrowed from these awesome source:
- [TORCHVISION OBJECT DETECTION FINETUNING TUTORIAL](https://pytorch.org/tutorials/intermediate/torchvision_tutorial.html)
- [TRANSFER LEARNING FOR COMPUTER VISION TUTORIAL](https://pytorch.org/tutorials/beginner/transfer_learning_tutorial.html)
- [Segmentation Models - Pytorch](https://github.com/qubvel/segmentation_models.pytorch)
- [Mask R-CNN for Object Detection and Segmentation](https://github.com/matterport/Mask_RCNN)
- [YOLOv5](https://github.com/ultralytics/yolov5)