SWIN transformer in Kaggle
Budget: $30 – $250 USD
Create a simple Kaggle notebook in Python that predicts instance segmentation for COCO dataset for a single class (that the user can choose) using SWIN-V2-base ("base" refers to the medium size SWIN model, which is the largest which is publicly available, and V2 means Version 2) model, and finally:
- Prints the mean Average Precision (mAP) on the whole class
- Picks a random image from the COCO class that has been selected by the user, and shows visually the segmentation predicted by the model
- Shows visually the bad segmentations (IoU <= 0.9) using Voxel51
Notice that SWIN-V2 weights are public, and there are many SWIN notebooks and notebooks that contain mAP available on Kaggle, so you only need to combine it to one Kaggle notebook.
Notice also that no user interface is required, so the user preferences can be nothing more than hard coded in a convenient place in the Kaggle notebook
- Prints the mean Average Precision (mAP) on the whole class
- Picks a random image from the COCO class that has been selected by the user, and shows visually the segmentation predicted by the model
- Shows visually the bad segmentations (IoU <= 0.9) using Voxel51
Notice that SWIN-V2 weights are public, and there are many SWIN notebooks and notebooks that contain mAP available on Kaggle, so you only need to combine it to one Kaggle notebook.
Notice also that no user interface is required, so the user preferences can be nothing more than hard coded in a convenient place in the Kaggle notebook