Update readme

README.md

@@ -1,20 +1,25 @@
-# Tiger detection
+# Tiger classification
 
-This repository contains tools to detect and classify tigers in camera trap images,
-using public camera trap images from LILA BC and free GPU resources in Google Colab.
+This repository contains scripts and notebooks to classify tigers (and other species) in camera trap images,
+using ML (e. g. [MegaDetector](https://github.com/agentmorris/MegaDetector)),
+open source tools and data (e. g. [LILA BC](https://lila.science/))
+and free compute resources (i. e. Colab and Kaggle).
 
-![tiger](anno_1440.jpg 'tiger')
+![tiger](media/anno_1440.jpg 'tiger')
 
 *Credentials: LILA BC, MegaDetector, own illustration.*
 
 
-## Motivation
+## Motivation and relevance
 
-- tigers are an endangered species
+- tigers are an endangered species,
+NGOs like the [Nepal Tiger Trust](https://www.nepaltigertrust.org/) protect them
 - there is no open and easy way for ecologists/researchers/NGOs
-to detect and classify their camera trap images with regard to tigers
+to classify their camera trap images with regard to tigers
 - ML and open data/tools can help reduce the amount of manual labor
-when sifting through large amounts of camera trap data, looking for the needle in the haystack
+when sifting through large amounts of camera trap images, looking for the needle in the haystack
+- goal: train a species classifier for Nepal (focussing on tigers)
+and make it available through [EcoAssist](https://addaxdatascience.com/ecoassist/)
 
 ## Data
 
@@ -23,20 +28,27 @@ when sifting through large amounts of camera trap data, looking for the needle i
 - [LILA BC](https://lila.science/)
 - amur tiger re-identification [challenge](https://cvwc2019.github.io/challenge.html) at CVWC 2019
 
-**Data preparation**
+**Sample and download images (Colab)**
 
-1. Define relevant species classes (tiger, other mamals, birds etc.)
-2. Sample n images per class randomly from LILA BC (e. g. from the last i years)
-3. Download sampled images and check for errors/inconsistencies
-4. Run images through [MegaDetector](https://github.com/agentmorris/MegaDetector) to get bounding boxes (and filter out empty images, vehicles and people if any)
-5. Use mewc-snip to crop images
-6. Build train, val and test sets
+1. Download image URLs and labels from LILA BC
+2. For each selected species: sample and download images, create train test split if applicable
+3. Copy images to Drive
+
+*Note: Since Colab and Drive have limited capacities, one might have to further split up the process.*
+
+**Preprocess images (Kaggle)**
+
+1. Run [MegaDetector](https://github.com/agentmorris/MegaDetector) on all images
+2. Snip images
+3. Copy snipped images to Kaggle Output
+
+*Note: Images must have been previously downloaded to Drive via Colab and then uploaded to Kaggle (zipped folder).*
 
 ## Training
 
 - [MEWC](https://github.com/zaandahl/mewc)
 - EfficientNetV2
-- Google Colab
+- Kaggle
 
 ## Deployment