clarify a few points and formatting

README.md

@@ -89,7 +89,7 @@ for more information on how this model can be used generate time-budgets from ae
 
 ### Training Data
 
-[KABR Dataset](https://huggingface.co/datasets/imageomics/KABR)
+This model was trained on the [KABR mini-scene dataset](https://huggingface.co/datasets/imageomics/KABR).
 
 ### Training Procedure 
 
@@ -103,7 +103,7 @@ For each tracklet, we create a separate video, called a mini-scene, by extractin
 detection in a video frame. 
 This allows us to compensate for the drone's movement and provides a stable, zoomed-in representation of the animal.
 
-See [project page](https://kabrdata.xyz/) and the [paper](https://openaccess.thecvf.com/content/WACV2024W/CV4Smalls/papers/Kholiavchenko_KABR_In-Situ_Dataset_for_Kenyan_Animal_Behavior_Recognition_From_Drone_WACVW_2024_paper.pdf) for data preprocessing details.
+See the [KBAR mini-scene project page](https://kabrdata.xyz/) and the [paper](https://openaccess.thecvf.com/content/WACV2024W/CV4Smalls/papers/Kholiavchenko_KABR_In-Situ_Dataset_for_Kenyan_Animal_Behavior_Recognition_From_Drone_WACVW_2024_paper.pdf) for data preprocessing details.
 
 We applied data augmentation techniques during training, including horizontal flipping to randomly
 mirror the input frames horizontally and color augmentations to randomly modify the
@@ -118,20 +118,17 @@ We used a sample rate of 16x5, and random weight initialization.
 
 ## Evaluation
 
-The dataset was evaluated on X3D-L model utilizing [SlowFast](https://github.com/facebookresearch/SlowFast) framework.
+The dataset was evaluated on the X3D-L model utilizing the [SlowFast](https://github.com/facebookresearch/SlowFast) framework, specifically utilizing teh [test_net script](https://github.com/facebookresearch/SlowFast/blob/main/tools/test_net.py).
 
-#### Testing Data
+### Testing Data
 
-[KABR Dataset](https://huggingface.co/datasets/imageomics/KABR)
-
-We provide a train-test split of the mini-scenes for evaluation purposes, with 75% for train and 25% for testing. No mini-scene was divided by the split. The splits ensured a stratified representation of giraffes, Plains zebras, and Grevy’s zebras.
+We provide a train-test split of the mini-scenes from the [KABR Dataset](https://huggingface.co/datasets/imageomics/KABR) for evaluation purposes (test set indicated in [annotations/val.csv](https://huggingface.co/datasets/imageomics/KABR/blob/main/KABR/annotation/val.csv), with 75% for train and 25% for testing. No mini-scene was divided by the split. The splits ensured a stratified representation of giraffes, Plains zebras, and Grevy’s zebras.
 
 #### Metrics
 
-We report precision, recall, and F1 score. We also report mean Average Precision (mAP) for overall, head-class, and tail-class performance.
-
+We report precision, recall, and F1 score on the KABR mini-scene test set, along with the mean Average Precision (mAP) for overall, head-class, and tail-class performance.
 
-### Results
+**Results**
 
 | WI       | BS | mAP Overall | mAP Head | mAP Tail | P      | R      | F1     |
 |----------|----|-------------|----------|----------|--------|--------|--------|
@@ -140,7 +137,7 @@ We report precision, recall, and F1 score. We also report mean Average Precision
 
 ### Model Architecture and Objective
 
-[Model Description](https://arxiv.org/pdf/2004.04730)
+Please see the [Base Model Description](https://arxiv.org/pdf/2004.04730).
 
 #### Hardware
 
@@ -150,7 +147,6 @@ Running the X3D model requires a modern NVIDIA GPU with CUDA support. X3D-L is d
 
 **BibTeX:**
 
-
 If you use our model in your work, please cite the model and associated paper.
 
 **Model**
@@ -207,5 +203,4 @@ Jenna Kline and Maksim Kholiavchenko
 
 ## Model Card Contact
 
-Maksim Kholiavchenko 
-<!-- Could include who to contact with questions, but this is also what the "Discussions" tab is for. -->
+For questions on this model, please open a [discussion](https://huggingface.co/imageomics/x3d-kabr-kinetics/discussions) on this repo.