at::Tensor to UIImage - pytorch

Question

I have a PyTorch model and try run it on iOS. I have the next code:
at::Tensor tensor2 = torch::from_blob(imageBuffer2, {1, 1, 256, 256}, at::kFloat);
c10::InferenceMode guard;
auto output = _impl.forward({tensor1, tensor2});
torch::Tensor tensor_img = output.toTuple()->elements()[0].toTensor();
My question is "How I can convert tensor_img to UIImage?"
I found that functions in PyTorch documentation:
- (UIImage*)convertRGBBufferToUIImage:(unsigned char*)buffer
withWidth:(int)width
withHeight:(int)height {
char* rgba = (char*)malloc(width * height * 4);
for (int i = 0; i < width * height; ++i) {
rgba[4 * i] = buffer[3 * i];
rgba[4 * i + 1] = buffer[3 * i + 1];
rgba[4 * i + 2] = buffer[3 * i + 2];
rgba[4 * i + 3] = 255;
}
size_t bufferLength = width * height * 4;
CGDataProviderRef provider = CGDataProviderCreateWithData(NULL, rgba, bufferLength, NULL);
size_t bitsPerComponent = 8;
size_t bitsPerPixel = 32;
size_t bytesPerRow = 4 * width;
CGColorSpaceRef colorSpaceRef = CGColorSpaceCreateDeviceRGB();
if (colorSpaceRef == NULL) {
NSLog(#"Error allocating color space");
CGDataProviderRelease(provider);
return nil;
}
CGBitmapInfo bitmapInfo = kCGBitmapByteOrderDefault | kCGImageAlphaPremultipliedLast;
CGColorRenderingIntent renderingIntent = kCGRenderingIntentDefault;
CGImageRef iref = CGImageCreate(width,
height,
bitsPerComponent,
bitsPerPixel,
bytesPerRow,
colorSpaceRef,
bitmapInfo,
provider,
NULL,
YES,
renderingIntent);
uint32_t* pixels = (uint32_t*)malloc(bufferLength);
if (pixels == NULL) {
NSLog(#"Error: Memory not allocated for bitmap");
CGDataProviderRelease(provider);
CGColorSpaceRelease(colorSpaceRef);
CGImageRelease(iref);
return nil;
}
CGContextRef context = CGBitmapContextCreate(pixels,
width,
height,
bitsPerComponent,
bytesPerRow,
colorSpaceRef,
bitmapInfo);
if (context == NULL) {
NSLog(#"Error context not created");
free(pixels);
}
UIImage* image = nil;
if (context) {
CGContextDrawImage(context, CGRectMake(0.0f, 0.0f, width, height), iref);
CGImageRef imageRef = CGBitmapContextCreateImage(context);
if ([UIImage respondsToSelector:#selector(imageWithCGImage:scale:orientation:)]) {
float scale = [[UIScreen mainScreen] scale];
image = [UIImage imageWithCGImage:imageRef scale:scale orientation:UIImageOrientationUp];
} else {
image = [UIImage imageWithCGImage:imageRef];
}
CGImageRelease(imageRef);
CGContextRelease(context);
}
CGColorSpaceRelease(colorSpaceRef);
CGImageRelease(iref);
CGDataProviderRelease(provider);
if (pixels) {
free(pixels);
}
return image;
}
#end
If I correctly understand, that function can convert unsigned char * to UIImage. I think that I need convert my tensor_img to unsigned char*, but I don't understand how I can do it.

Answer 1

The 1st code its a torch bridge and 2nd code is UIImage helper which I run from Swift. Anyway, I resolve that issue, we can close it. Code example:
for (int i = 0; i < 3 * width * height; i++) {
[results addObject:#(floatBuffer[i])];
}
NSMutableData* data = [NSMutableData dataWithLength:sizeof(float) * 3 * width * height];
float* buffer = (float*)[data mutableBytes];
for (int j = 0; j < 3 * width * height; j++) {
buffer[j] = [results[j] floatValue];
}
return buffer;