Processing Pipeline

Chloros 1.1.0 uses a 4-thread processing pipeline that operates as a staged assembly line. Each thread handles a distinct phase of the processing workflow, allowing multiple images to be processed concurrently at different stages.

---

Pipeline Architecture

Images In → [Thread 1: Detection] → [Thread 2: Calibration] → [Thread 3: Processing] → [Thread 4: Export] → Files Out

Each image flows through all four threads in order. With Chloros+ multi-threaded processing, multiple images can be in different threads simultaneously — while Thread 3 processes one image, Thread 1 can be detecting the next, Thread 2 can be calibrating another, and Thread 4 can be writing a previously processed image to disk.

---

Thread Details

Thread 1: Detection

Purpose: Load images and detect calibration targets.

• Reads image files from disk (RAW, JPG)

• Extracts EXIF metadata (GPS, camera model, timestamps, exposure)

• Detects ArUco calibration targets in marked target images

• Outputs: image data + metadata + target detection results

This is primarily an I/O and CPU-bound thread.

Thread 2: Calibration

Purpose: Compute calibration parameters from detected targets.

• Calculates reflectance calibration coefficients from target images

• Computes vignette correction parameters

• Determines per-band calibration curves

• Outputs: calibration parameters for each image

This is a CPU-bound computation thread.

Thread 3: Processing (GPU)

Purpose: Apply corrections and calculate vegetation indices. This is the most compute-intensive thread.

• Debayering: Converts RAW Bayer pattern data to multi-channel images

  • Standard (Fast, Medium Quality) — default

  • Texture Aware (Slow, Highest Quality) — Chloros+ only, uses AI/ML denoising

• Vignette correction: Applies lens vignette correction across the image

• Reflectance calibration: Applies calibration coefficients to convert to reflectance values

• Index calculation: Computes vegetation indices (NDVI, NDRE, GNDVI, etc.)

• Outputs: processed image data ready for export

This thread benefits most from GPU acceleration. The Dynamic Compute Adaptation system primarily optimizes this thread's behavior.

Thread 4: Export

Purpose: Write processed images to disk.

• Writes output files in the selected format (TIFF 16-bit, TIFF 32-bit %, PNG, JPG)

• Embeds EXIF metadata in output files (GPS, timestamps, processing parameters)

• Organizes output into camera-model subfolders

• Outputs: final files on disk

This is primarily an I/O-bound thread. SSD storage significantly improves Thread 4 performance.

---

Sequential vs. Pipelined Processing

Free Mode (Sequential)

In the free version of Chloros, images are processed one at a time, sequentially through all four stages:

Image 1: [Detect] → [Calibrate] → [Process] → [Export]
                                                         Image 2: [Detect] → [Calibrate] → [Process] → [Export]

The GUI progress bar shows 2 stages: Target Detect and Processing.

Chloros+ Mode (Pipelined)

With a Chloros+ license, all four threads operate concurrently on different images:

Thread 1: [Image 1] [Image 2] [Image 3] [Image 4] ...
Thread 2:           [Image 1] [Image 2] [Image 3] ...
Thread 3:                     [Image 1] [Image 2] ...
Thread 4:                               [Image 1] ...

The GUI progress bar shows 4 stages: Detecting, Analyzing, Calibrating, Exporting. Hover over the progress bar to see per-thread progress.

Pipelined processing with Chloros+ can be 3-5x faster than sequential processing, depending on your hardware and dataset size. The speedup is greatest on systems with fast GPUs and SSDs.

---

Thread 4 Export Progress

In Chloros 1.1.0, the export thread (Thread 4) has its own dedicated progress tracking. You can monitor export progress separately:

CLI:

chloros-cli export-status

SDK:

status = chloros.get_status()
print(f"Export: {status['export']['percent']}% - Phase: {status['export']['phase']}")

Processing is complete when Thread 4 reaches 100%.

---

Relationship to Dynamic Compute Adaptation

The Dynamic Compute Adaptation system primarily affects Thread 3 (Processing):

• GPU_PARALLEL strategy: Thread 3 runs multiple images through the GPU simultaneously using the fused_gpu pipeline

• GPU_SINGLE strategy: Thread 3 processes one image at a time using the memory-efficient tiled_gpu pipeline

• CPU_PARALLEL strategy: Thread 3 uses CPU-based processing with multi-threaded parallelism

Thread 3's GPU memory allocation also changes dynamically as Threads 1 and 2 complete — see Dynamic GPU Memory Allocation.

---

Next Steps

• Dynamic Compute Adaptation — How Chloros selects the optimal strategy for your hardware

• NVIDIA Jetson Guide — Platform-specific pipeline behavior on Jetson

• Monitoring the Processing — GUI progress monitoring