diff options
| -rw-r--r-- | src/client/views/nodes/ImageBox.tsx | 83 | ||||
| -rw-r--r-- | src/client/views/search/FaceRecognitionHandler.tsx | 5 |
2 files changed, 34 insertions, 54 deletions
diff --git a/src/client/views/nodes/ImageBox.tsx b/src/client/views/nodes/ImageBox.tsx index 1e16bbfc9..d7e21b0a6 100644 --- a/src/client/views/nodes/ImageBox.tsx +++ b/src/client/views/nodes/ImageBox.tsx @@ -142,56 +142,39 @@ export class ImageBox extends ViewBoxAnnotatableComponent<FieldViewProps>() { }; autoTag = async () => { - - try { - // 1) grab the full-size URL - const layoutKey = Doc.LayoutDataKey(this.Document); - const url = ImageCastWithSuffix(this.Document[layoutKey], '_o') ?? ''; - if (!url) throw new Error('No image URL found'); - - // 2) convert to base64 - const base64 = await imageUrlToBase64(url); - if (!base64) throw new Error('Failed to load image data'); - - // 3) ask GPT for labels one label: PERSON or LANDSCAPE - const raw = await gptImageLabel( - base64, - `Classify this image as PERSON or LANDSCAPE. You may only respond with one of these two options. Then - provide five additional descriptive tags to describe the image for a total of 6 words outputted, - delimited by spaces. For example: "LANDSCAPE BUNNY NATURE FOREST PEACEFUL OUTDOORS". Then add one final lengthier summary tag (separated by underscores) - that describes the image.` - ); - - const { nativeWidth, nativeHeight } = this.nativeSize; - const aspectRatio = nativeWidth && nativeHeight - ? (nativeWidth / nativeHeight).toFixed(2) - : '1.00'; - - // 4) normalize and prefix - const label = raw - .trim() - .toUpperCase() - - // 5) stash it on the Doc - // overwrite any old tags so re-runs still work - const tokens = label.split(/\s+/); - this.Document.$tags_chat = new List<string>(); - tokens.forEach(tok => { - (this.Document.$tags_chat as List<string>).push(tok)}); - (this.Document.$tags_chat as List<string>).push(`ASPECT_${aspectRatio}`); - - // 6) flip on “show tags” in the layout - // (same flag that ImageLabelBox.toggleDisplayInformation uses) - this.Document._layout_showTags = true; - - } catch (err) { - console.error('autoTag failed:', err); - } finally { - } - }; - - - + if (this.Document.$tags_chat) return; + try { + // 1) grab the full-size URL + const layoutKey = Doc.LayoutDataKey(this.Document); + const url = ImageCastWithSuffix(this.Document[layoutKey], '_o'); + if (!url) throw new Error('No image URL found'); + + // 2) convert to base64 + const base64 = await imageUrlToBase64(url); + if (!base64) throw new Error('Failed to load image data'); + + // 3) ask GPT for labels one label: PERSON or LANDSCAPE + const label = await gptImageLabel( + base64, + `Classify this image as PERSON or LANDSCAPE. You may only respond with one of these two options. + Then provide five additional descriptive tags to describe the image for a total of 6 words outputted, delimited by spaces. + For example: "LANDSCAPE BUNNY NATURE FOREST PEACEFUL OUTDOORS". + Then add one final lengthier summary tag (separated by underscores) that describes the image.` + ).then(raw => raw.trim().toUpperCase()); + + const { nativeWidth, nativeHeight } = this.nativeSize; + const aspectRatio = ((nativeWidth || 1) / (nativeHeight || 1)).toFixed(2); + + // 5) stash it on the Doc + // overwrite any old tags so re-runs still work + this.Document.$tags_chat = new List<string>([...label.split(/\s+/), `ASPECT_${aspectRatio}`]); + + // 6) flip on “show tags” in the layout + this.Document._layout_showTags = true; + } catch (err) { + console.error('autoTag failed:', err); + } + }; getAnchor = (addAsAnnotation: boolean, pinProps?: PinProps) => { const visibleAnchor = this._getAnchor?.(this._savedAnnotations, true); // use marquee anchor, otherwise, save zoom/pan as anchor diff --git a/src/client/views/search/FaceRecognitionHandler.tsx b/src/client/views/search/FaceRecognitionHandler.tsx index 256e68afd..dac91b89a 100644 --- a/src/client/views/search/FaceRecognitionHandler.tsx +++ b/src/client/views/search/FaceRecognitionHandler.tsx @@ -9,7 +9,6 @@ import { ImageField } from '../../../fields/URLField'; import { DocumentType } from '../../documents/DocumentTypes'; import { Docs } from '../../documents/Documents'; import { DocumentManager } from '../../util/DocumentManager'; -import { reaction } from 'mobx'; import { DocumentView } from '../nodes/DocumentView'; /** @@ -210,9 +209,7 @@ export class FaceRecognitionHandler { } else if (imgDoc.type === DocumentType.LOADING && !imgDoc.loadingError) { setTimeout(() => this.classifyFacesInImage(imgDocView), 1000); } else { - reaction(() => ({sel:imgDocView.isSelected()}), ({sel}) => !sel && - imgDocView.ComponentView?.autoTag?.(), {fireImmediately: true} - ) + imgDocView.ComponentView?.autoTag?.(); const imgUrl = ImageCast(imgDoc[Doc.LayoutDataKey(imgDoc)]); if (imgUrl && !DocListCast(Doc.MyFaceCollection?.examinedFaceDocs).includes(imgDoc[DocData])) { // only examine Docs that have an image and that haven't already been examined. |