tutorials GitHub

Section 7: Prompt Engineering and Generating Image Detection Code

Lecture Slides

FlexServ API Prompt: YOLO Evaluation Script Generator

Task Summary:

To test the capabilities of the FlexServ inference server, we can provide a complex prompt to the Responses API. This prompt asks the AI to generate a complete Python evaluation script that performs Animal detection on the images from the LILA BC Small Animal dataset. This is a large camera-trap image dataset used for wildlife monitoring and ecological research. It contains millions of images captured by automated cameras, including small mammals and many blank triggers, along with annotations describing the detected species. For training object detection models such as YOLO, the dataset can be downloaded in YOLO format, where each image has a corresponding .txt label file containing bounding-box coordinates in the form .

Exploring the FlexServ UI

Step 7.1: Refresh Model Pool

  • Refresh the Model pool so you can see public and private models available for you to run. Model Pool

Step 7.2: Update the Responses API and Parameters

  • Copy and paste the following prompt into the FlexServ UI in the Responses API, Input(Markdown) section, shown in the image below.

TASK DESCRIPTION:
This is an IMAGE-LEVEL BINARY CLASSIFICATION task implemented using an object detection model.
The goal is to determine whether an image contains an animal or not.

DATASET STRUCTURE:
DATASET_ROOT contains three subdirectories: train, test, and val.
Each directory contains two subdirectories:
images/ → contains image files (.jpg, .jpeg, .png)
labels/ → contains YOLO format .txt files

GROUND-TRUTH LOGIC: An image is considered an animal if a corresponding .txt file exists and is not empty in the labels/ folder.

MODEL REQUIREMENTS:
Use ONLY a pretrained Ultralytics YOLO detection model (e.g., yolov8n.pt).
Load the model using the Ultralytics YOLO API.
Assume YOLO detects animals using class ID animal at index 0.

DETECTION LOGIC (IMPORTANT):
Run object detection on each image.
If the model produces AT LEAST ONE detection of an animal class with confidence >= 0.5:
→ The image-level prediction is animal.

EVALUATION METRICS:
Iterate through the images in the test split.
Compare the image-level prediction with the ground truth (existence of label file).
Count: True Positives, True Negatives, False Positives, and False Negatives.

ACCURACY DEFINITION:
Overall accuracy = (True Positives + True Negatives) / Total Images

OUTPUT REQUIREMENTS:
Print for each image: filename, ground-truth status, and prediction.
At the end, print a summary report including total images, counts for each metric, and overall detection accuracy.

CODING REQUIREMENTS:
Store the main path in DATASET_ROOT.
Use pathlib or os for robust file path matching.
Read only .jpg files.
Include clear comments explaining each step.

After the code, briefly explain how the program works in plain English.

Paste Prompt

  • Change the temperature to a value 0.0 for a deterministic solution.
  • Select the model to run
    • Qwen/Qwen2.5-Coder32B-Instruct-61.0 GB - Text Generation
  • Make sure the Streams is checked.
  • Uncheck Multi-turn conversation

### Step 7.3: Run the Responses API

  • Click Run. Within a few minutes, you should see the code generation start in the blue box in the Responses API. Wait for it to complete. After completion, you should see output similar to the image below.

Code

Step 7.4: Running Code Detection On Jupyter

Go to the Jupyter notebook Code-Detection on your Jupyter path. ai-tutorial-2026 -> notebooks -> Code-Detection.ipynb

Copy the generated code from FlexServ UI in a new cell below the cell titled Put your generated code here.

Update the variable DATASET_ROOT to path /home/jovyan/ai-tutorial-2026/datasets/AnimalEcology.v4i.yolov11 in your generated code

Also update the model path to /home/jovyan/ai-tutorial-2026/models/yolov9t_ep200_bs32_lr0.005_baa22147.pt

Now run the code. On successful run, you should see output similar to below


image 1/1 /home/jovyan/work/vista/ai-tutorial-2026/datasets/AnimalEcology.v4i.yolov11/test/images/KPC2__2019-09-19__15-47-42-1-_JPG.rf.608031a2809f0f6714f175d3e5eb7f06.jpg: 640x640 1 animal, 96.6ms
Filename: KPC2__2019-09-19__15-47-42-1-_JPG.rf.608031a2809f0f6714f175d3e5eb7f06.jpg, Ground Truth: no_animal, Prediction: animal

Speed: 2.5ms preprocess, 96.6ms inference, 1.0ms postprocess per image at shape (1, 3, 640, 640)
image 1/1 /home/jovyan/work/vista/ai-tutorial-2026/datasets/AnimalEcology.v4i.yolov11/test/images/NOR3__2019-07-19__11-40-00-1-_JPG.rf.b85ee30f99a803b09f8c5a7da7f9a508.jpg: 640x640 (no detections), 104.2ms
Speed: 1.9ms preprocess, 104.2ms inference, 0.7ms postprocess per image at shape (1, 3, 640, 640)
Filename: NOR3__2019-07-19__11-40-00-1-_JPG.rf.b85ee30f99a803b09f8c5a7da7f9a508.jpg, Ground Truth: animal, Prediction: no_animal
....
...
Evaluation Metrics:
Total images processed: 100
Total animal images (based on label files): 71
True Positives: 47
True Negatives: 6
False Positives: 23
False Negatives: 24
Overall detection accuracy: 0.53