update readme to be good

README.md

@@ -1,4 +1,4 @@
-```md
+
 ---
 license: mit
 library_name: ultralytics
@@ -19,21 +19,9 @@ datasets:
   - garythung/trashnet
 ---
 
-# MIRA — Machine Intelligence for Recycling Automation
-
-MIRA is a computer-vision project for recognizing waste materials and preparing
-the results for future automated sorting.
-
-The project was developed for Jugend forscht. The long-term goal is to connect
-the vision system to a robot or sorting mechanism. Raspberry Pi deployment and
-physical sorting are still future work.
-
-- GitHub: https://github.com/jeremy341/MIRA-AI
-- Project website: https://mira-vision.vercel.app/
+# MIRA — Waste Detection Models
 
-## What the models detect
-
-All detection models use five classes:
+MIRA is a computer-vision project for detecting five types of waste:
 
 - glass
 - metal
@@ -41,36 +29,24 @@ All detection models use five classes:
 - plastic
 - trash
 
-## Model files
-
-This repository contains models and exports from several experiments:
-
-| File type | Description |
-|---|---|
-| `.pt` | Ultralytics YOLO/PyTorch models |
-| `.onnx` | ONNX exports |
-| `.tflite` | Quantized LiteRT/TFLite exports |
+This repository contains the trained models and exports from the MIRA
+experiments. The recommended model is `mira_exp019.pt`.
 
-The recommended current detector is:
-
-```text
-mira_exp019.pt
-```
+- GitHub: https://github.com/jeremy341/MIRA-AI
+- Project website: https://mira-vision.vercel.app/
 
-The EXP-019 exports are:
+## Recommended model
 
-| File | Description |
-|---|---|
-| `mira_exp019.pt` | Main YOLO11n detector |
-| `mira_exp019.onnx` | ONNX export |
-| `mira_exp019_int8_320.tflite` | INT8 TFLite export at 320 px |
-| `mira_exp019_int8_640.tflite` | INT8 TFLite export at 640 px |
+| File | Format | Description |
+|---|---|---|
+| `mira_exp019.pt` | PyTorch | Recommended YOLO11n detector |
+| `mira_exp019.onnx` | ONNX | ONNX export of EXP-019 |
+| `mira_exp019_int8_320.tflite` | TFLite | INT8 export at 320 px |
+| `mira_exp019_int8_640.tflite` | TFLite | INT8 export at 640 px |
 
 ## Experiment results
 
-The main detection experiments produced the following results:
-
-| Experiment | Model | Dataset setup | mAP50 |
+| Experiment | Model | Dataset | mAP50 |
 |---|---|---|---:|
 | EXP-005 | YOLOv8n | Custom + TrashNet | 82.3% |
 | EXP-006 | YOLOv8n | Fused Wild + TrashNet | 39.4% |
@@ -84,17 +60,11 @@ The main detection experiments produced the following results:
 | EXP-018 | YOLO11n | Clean balanced dataset | 90.6% |
 | EXP-019 | YOLO11n | Clean balanced repeatability run | 90.58% |
 
-The main result was that adding more data did not automatically improve the
-model. Some larger combinations contained inconsistent image styles, labels,
-and object arrangements.
+The main finding was that more data did not automatically produce better
+results. Removing inconsistent data and building a cleaner, balanced dataset
+led to the strongest results in EXP-018 and EXP-019.
 
-After removing unsuitable data and creating a cleaner, more balanced dataset,
-the detector improved substantially. EXP-019 repeated the result from EXP-018
-closely.
-
-## EXP-019 evaluation
-
-The recorded EXP-019 metrics are:
+## EXP-019 metrics
 
 | Metric | Result |
 |---|---:|
@@ -105,77 +75,51 @@ The recorded EXP-019 metrics are:
 | Training images | 5,108 |
 | Validation images | 415 |
 | Test images | 1,375 |
-| Training time | 2.672 hours |
-| Hardware | NVIDIA Tesla T4 |
 
-These results come from the project’s validation and evaluation setup. They
-should not be interpreted as completed real-world or robotic-sorting
-performance.
+These results come from the project’s recorded evaluation setup.
 
 ## Usage
 
-Download a model file and install Ultralytics:
+Install Ultralytics:
 
 ```bash
 pip install ultralytics
 ```
 
-Load the model and run prediction:
+Download `mira_exp019.pt` from this repository and run inference:
 
 ```python
 from ultralytics import YOLO
 
 model = YOLO("mira_exp019.pt")
-
-results = model.predict(
-    source="image.jpg",
-    conf=0.25,
-    save=True,
-)
+results = model.predict("image.jpg", conf=0.25, save=True)
 ```
 
-For validation, a compatible YOLO dataset YAML is required:
+For validation, provide a compatible YOLO dataset configuration:
 
 ```python
 results = model.val(data="dataset.yaml")
 ```
 
-The MIRA repository provides the full CLI, webcam interface, dashboard,
-dataset tools, and download commands:
-
-```powershell
-mira download mira_exp019.pt
-mira live --model mira_exp019.pt
-mira dashboard
-```
-
 ## Datasets
 
-The project used combinations of:
+The models were trained using combinations of:
 
 - [dmedhi garbage image classification/detection](https://huggingface.co/datasets/dmedhi/garbage-image-classification-detection)
 - [TACO](https://github.com/pedropro/TACO)
 - [TrashNet](https://github.com/garythung/trashnet)
 - Roboflow waste-detection data
 
-The datasets were remapped to the five MIRA classes. Dataset licenses and
-source details are documented in the
-[GitHub repository](https://github.com/jeremy341/MIRA-AI/blob/main/docs/DATASET_ORIGINS.md).
+The datasets were remapped to the five MIRA classes. Their original licenses
+and sources remain applicable.
 
 ## Limitations
 
-The current models can struggle with:
-
-- white crumpled paper
-- cans viewed from the opening
-- strongly overlapping objects
-- unusual lighting and object arrangements
-
-Raspberry Pi latency, memory use, and complete physical sorting have not yet
-been fully tested.
+The models can struggle with crumpled paper, cans viewed from the opening,
+overlapping objects, and unusual image conditions.
 
 ## License
 
-The project code and model files are released under the MIT License where
-applicable. Dataset licenses remain subject to their original terms.
+The model files are provided under the MIT License where applicable. Dataset
+licenses remain subject to their original terms.
 ```