Solution For kaggle challange for iWildCam 2019
- Used TF example
- Dataset was downloaded and extracted
- Trained over this dataset
- Used it on our data (inference then train / fine-tune)
- Split to train / validate
- Used spacial feature extraction
- Evaluate on train / validate / test set
- Submission with dummy file, just to check the format
- Submit to Kaggl
Go to Kaggle page.
Download iwildcam-2019-fgvc6.zip
file (46.6G).
Using TensorFlow 2.1, and tensorflow_hub. Other pakedges:
- seaborn
- matplotlib
TF hub page: link.
python train_val.py
Change the run
variable to the current run folder.
Set dataset to the desired one.
The results (CSV file, for submition) should be at the run folder.
python evaluate.py
Go to submission page at Kaggle and drop the CSV file.