File Conversion for Synapse Upload

Knowledge about when and how to convert files before uploading to a data collection.

When Conversion Is Needed

Conversion is required when source file extensions don't match the data collection's file specification allowed extensions.

Example: Spec image_1 allows [".png", ".jpg"] but the source directory contains .tiff files. The TIFFs must be converted to PNG or JPG before upload.

Detection Logic

1. Get the file specs from the data collection (each spec has extensions list)
2. Scan the source directory for actual file extensions
3. For each spec, check if source extensions are in the allowed list
4. If not, determine if a supported conversion exists

Supported Conversions

Image Conversions (via Pillow)

Source	Target	Notes
.tiff, .tif	.png	Lossless conversion
.tiff, .tif	.jpg, .jpeg	RGBA/P modes auto-converted to RGB
.bmp	.png	Lossless conversion
.bmp	.jpg, .jpeg	Mode conversion applied
.webp	.png	Lossless conversion
.webp	.jpg, .jpeg	Mode conversion applied
.gif	.png	First frame only

Requirement: Pillow>=10.0 must be installed in the upload plugin environment.

Video Conversions (via ffmpeg)

Source	Target	Notes
.mov	.mp4	H.264 + AAC, faststart
.avi	.mp4	H.264 + AAC, faststart
.mkv	.mp4	H.264 + AAC, faststart
.webm	.mp4	H.264 + AAC, faststart
.flv	.mp4	H.264 + AAC, faststart
.wmv	.mp4	H.264 + AAC, faststart

Requirement: ffmpeg must be available in PATH. Timeout: 600 seconds per file.

Audio Conversions (via ffmpeg)

Source	Target	Notes
.wav	.mp3	Default quality
.flac	.mp3	Lossy conversion
.ogg	.mp3	Lossy conversion
.m4a	.mp3	Lossy conversion
.wma	.mp3	Lossy conversion
.wav	.aac	Default quality
.flac	.aac	Lossy conversion
.ogg	.aac	Lossy conversion

Requirement: ffmpeg must be available in PATH. Timeout: 300 seconds per file.

Conversion in the Upload Workflow

Using the ai-upload-plugin's Built-in Conversion

The ai-upload-plugin has a convert_file tool that handles conversion automatically:

{
  "name": "convert_file",
  "input": {
    "source_path": "/data/patient_001/scan.tiff",
    "target_format": ".png"
  }
}

Converted files are stored in a temp directory and automatically cleaned up after upload.

Manual Conversion Before Upload

If not using the ai-upload-plugin, convert files beforehand:

Images (Pillow):

python3 -c "
from PIL import Image
from pathlib import Path
import sys

src = Path(sys.argv[1])
dst = src.with_suffix('.png')
with Image.open(src) as img:
    img.save(dst)
print(f'Converted: {dst}')
" /path/to/image.tiff

Batch image conversion:

python3 -c "
from PIL import Image
from pathlib import Path
import sys

src_dir = Path(sys.argv[1])
for tiff in src_dir.rglob('*.tiff'):
    dst = tiff.with_suffix('.png')
    with Image.open(tiff) as img:
        img.save(dst)
    print(f'Converted: {dst}')
" /path/to/directory

Video (ffmpeg):

ffmpeg -i input.mov -c:v libx264 -c:a aac -movflags +faststart output.mp4

Audio (ffmpeg):

ffmpeg -i input.wav output.mp3

Conversion Decision Tree

Source file extension NOT in spec's allowed extensions?
├── YES → Check if a supported conversion exists
│   ├── YES → Convert the file, use converted path for upload
│   └── NO → Report error: unsupported file format
└── NO → No conversion needed, use source file directly

Important Notes

• Converted files use a temporary directory that is cleaned up after the upload completes
• For large datasets, conversion can be the bottleneck — consider pre-converting files outside the upload workflow
• Image conversion preserves the original filename stem, only changing the extension
• RGBA/Palette mode images are automatically converted to RGB when targeting JPEG format
• If ffmpeg is not installed, video/audio conversions will fail gracefully with a warning