ente/lib/services/semantic_search/frameworks/ml_framework.dart
2023-12-13 14:37:26 +05:30

126 lines
3.9 KiB
Dart

import "dart:io";
import "package:flutter/services.dart";
import "package:logging/logging.dart";
import "package:path/path.dart";
import "package:path_provider/path_provider.dart";
import "package:photos/core/network/network.dart";
abstract class MLFramework {
final _logger = Logger("MLFramework");
/// Returns the path of the Image Model hosted remotely
String getImageModelRemotePath();
/// Returns the path of the Text Model hosted remotely
String getTextModelRemotePath();
/// Loads the Image Model stored at [path] into the framework
Future<void> loadImageModel(String path);
/// Loads the Text Model stored at [path] into the framework
Future<void> loadTextModel(String path);
/// Returns the Image Embedding for a file stored at [imagePath]
Future<List<double>> getImageEmbedding(String imagePath);
/// Returns the Text Embedding for [text]
Future<List<double>> getTextEmbedding(String text);
/// Downloads the models from remote, caches them and loads them into the
/// framework. Override this method if you would like to control the
/// initialization. For eg. if you wish to load the model from `/assets`
/// instead of a CDN.
Future<void> init() async {
await _initImageModel();
await _initTextModel();
}
/// Returns the cosine similarity between [imageEmbedding] and [textEmbedding]
double computeScore(List<double> imageEmbedding, List<double> textEmbedding) {
assert(
imageEmbedding.length == textEmbedding.length,
"The two embeddings should have the same length",
);
double score = 0;
for (int index = 0; index < imageEmbedding.length; index++) {
score += imageEmbedding[index] * textEmbedding[index];
}
return score;
}
// ---
// Private methods
// ---
Future<void> _initImageModel() async {
const assetPath = "assets/models/clip/clip-image-vit-32-float32.onnx";
await loadImageModel(
await getAccessiblePathForAsset(
assetPath,
"clip-image-vit-32-float32.onnx",
),
);
//final path = await _getLocalImageModelPath();
// if (File(path).existsSync()) {
// await loadImageModel(path);
// } else {
// final tempFile = File(path + ".temp");
// await _downloadFile(getImageModelRemotePath(), tempFile.path);
// await tempFile.rename(path);
// await loadImageModel(path);
// }
}
Future<void> _initTextModel() async {
const assetPath = "assets/models/clip/clip-text-vit-32-int32.onnx";
await loadTextModel(
await getAccessiblePathForAsset(
assetPath,
"clip-text-vit-32-float32.onnx",
),
);
//final path = await _getLocalTextModelPath();
// if (File(path).existsSync()) {
// await loadTextModel(path);
// } else {
// final tempFile = File(path + ".temp");
// await _downloadFile(getTextModelRemotePath(), tempFile.path);
// await tempFile.rename(path);
// await loadTextModel(path);
// }
}
Future<String> _getLocalImageModelPath() async {
return (await getTemporaryDirectory()).path +
"/models/" +
basename(getImageModelRemotePath());
}
Future<String> _getLocalTextModelPath() async {
return (await getTemporaryDirectory()).path +
"/models/" +
basename(getTextModelRemotePath());
}
Future<void> _downloadFile(String url, String savePath) async {
_logger.info("Downloading " + url);
final existingFile = File(savePath);
if (await existingFile.exists()) {
await existingFile.delete();
}
await NetworkClient.instance.getDio().download(url, savePath);
}
Future<String> getAccessiblePathForAsset(
String assetPath,
String tempName,
) async {
final byteData = await rootBundle.load(assetPath);
final tempDir = await getTemporaryDirectory();
final file = await File('${tempDir.path}/$tempName')
.writeAsBytes(byteData.buffer.asUint8List());
return file.path;
}
}