Use WEBP to encode thumbnails

sist2app · May 20, 2023 · 6108821 · 6108821
1 parent e2e0cf2
commit 6108821
Show file tree

Hide file tree

Showing 10 changed files with 73 additions and 46 deletions.
diff --git a/README.md b/README.md
@@ -28,7 +28,7 @@ sist2 (Simple incremental search tool)
 
 \* See [format support](#format-support)    
 \*\* See [Archive files](#archive-files)    
-\*\*\* See [OCR](#ocr)
+\*\*\* See [OCR](#ocr)    
 \*\*\*\* See [Named-Entity Recognition](#NER)
 
 ## Getting Started
@@ -46,7 +46,7 @@ services:
       - "discovery.type=single-node"
       - "ES_JAVA_OPTS=-Xms2g -Xmx2g"
   sist2-admin:
-    image: simon987/sist2:3.0.7-x64-linux
+    image: simon987/sist2:3.1.0-x64-linux
     restart: unless-stopped
     volumes:
       - ./sist2-admin-data/:/sist2-admin/
@@ -206,7 +206,7 @@ docker run --rm --entrypoint cat my-sist2-image /root/sist2 > sist2-x64-linux
 3. Install vcpkg dependencies
 
     ```bash
-    vcpkg install curl[core,openssl] sqlite3[core,fts5] cpp-jwt pcre cjson brotli libarchive[core,bzip2,libxml2,lz4,lzma,lzo] pthread tesseract libxml2 libmupdf gtest mongoose libmagic libraw gumbo ffmpeg[core,avcodec,avformat,swscale,swresample]
+    vcpkg install curl[core,openssl] sqlite3[core,fts5] cpp-jwt pcre cjson brotli libarchive[core,bzip2,libxml2,lz4,lzma,lzo] pthread tesseract libxml2 libmupdf gtest mongoose libmagic libraw gumbo ffmpeg[core,avcodec,avformat,swscale,swresample,webp]
     ```
 
 4. Build

diff --git a/docs/USAGE.md b/docs/USAGE.md
@@ -17,7 +17,7 @@ Lightning-fast file system indexer and search tool.
 
 Scan options
     -t, --threads=<int>               Number of threads. DEFAULT: 1
-    -q, --thumbnail-quality=<int>     Thumbnail quality, on a scale of 2 to 31, 2 being the best. DEFAULT: 2
+    -q, --thumbnail-quality=<int>     Thumbnail quality, on a scale of 0 to 100, 100 being the best. DEFAULT: 50
     --thumbnail-size=<int>            Thumbnail size, in pixels. DEFAULT: 552
     --thumbnail-count=<int>           Number of thumbnails to generate. Set a value > 1 to create video previews, set to 0 to disable thumbnails. DEFAULT: 1
     --content-size=<int>              Number of bytes to be extracted from text documents. Set to 0 to disable. DEFAULT: 32768
@@ -88,8 +88,8 @@ Made by simon987 <[email protected]>. Released under GPL-3.0
 
 See chart below for rough estimate of thumbnail size vs. thumbnail size & quality arguments:
 
-For example, `--thumbnail-size=500`, `--thumbnail-quality=2` for a directory with 8 million images will create a thumbnail database 
-that is about `8000000 * 36kB = 288GB`.
+For example, `--thumbnail-size=500`, `--thumbnail-quality=50` for a directory with 8 million images will create a thumbnail database 
+that is about `8000000 * 11.8kB = 94.4GB`.
 
 ![thumbnail_size](thumbnail_size.png)
 

diff --git a/docs/thumbnail_size.png b/docs/thumbnail_size.png
diff --git a/src/cli.c b/src/cli.c
@@ -5,7 +5,7 @@
 #define DEFAULT_OUTPUT "index.sist2"
 #define DEFAULT_NAME "index"
 #define DEFAULT_CONTENT_SIZE 32768
-#define DEFAULT_QUALITY 2
+#define DEFAULT_QUALITY 50
 #define DEFAULT_THUMBNAIL_SIZE 552
 #define DEFAULT_THUMBNAIL_COUNT 1
 #define DEFAULT_REWRITE_URL ""
@@ -100,16 +100,16 @@ int scan_args_validate(scan_args_t *args, int argc, const char **argv) {
 
     if (args->tn_quality == OPTION_VALUE_UNSPECIFIED) {
         args->tn_quality = DEFAULT_QUALITY;
-    } else if (args->tn_quality < 2 || args->tn_quality > 31) {
-        fprintf(stderr, "Invalid value for --thumbnail-quality argument: %d. Must be within [2, 31].\n",
+    } else if (args->tn_quality < 0 || args->tn_quality > 100) {
+        fprintf(stderr, "Invalid value for --thumbnail-quality argument: %d. Must be within [0, 100].\n",
                 args->tn_quality);
         return 1;
     }
 
     if (args->tn_size == OPTION_VALUE_UNSPECIFIED) {
         args->tn_size = DEFAULT_THUMBNAIL_SIZE;
     } else if (args->tn_size < 32) {
-        printf("Invalid value --thumbnail-size argument: %d. Must be greater than 32 pixels.\n", args->tn_size);
+        printf("Invalid value --thumbnail-size argument: %d. Must be >= 32 pixels.\n", args->tn_size);
         return 1;
     }
 

diff --git a/src/main.c b/src/main.c
@@ -490,7 +490,7 @@ int main(int argc, const char *argv[]) {
             OPT_GROUP("Scan options"),
             OPT_INTEGER('t', "threads", &common_threads, "Number of threads. DEFAULT: 1"),
             OPT_INTEGER('q', "thumbnail-quality", &scan_args->tn_quality,
-                        "Thumbnail quality, on a scale of 2 to 31, 2 being the best. DEFAULT: 2",
+                        "Thumbnail quality, on a scale of 0 to 100, 100 being the best. DEFAULT: 50",
                         set_to_negative_if_value_is_zero, (intptr_t) &scan_args->tn_quality),
             OPT_INTEGER(0, "thumbnail-size", &scan_args->tn_size,
                         "Thumbnail size, in pixels. DEFAULT: 552",

diff --git a/src/sist.h b/src/sist.h
@@ -51,11 +51,11 @@
 #include <ctype.h>
 #include "git_hash.h"
 
-#define VERSION "3.0.7"
+#define VERSION "3.1.0"
 static const char *const Version = VERSION;
 static const int VersionMajor = 3;
-static const int VersionMinor = 0;
-static const int VersionPatch = 7;
+static const int VersionMinor = 1;
+static const int VersionPatch = 0;
 
 #ifndef SIST_PLATFORM
 #define SIST_PLATFORM unknown

diff --git a/third-party/libscan/libscan/ebook/ebook.c b/third-party/libscan/libscan/ebook/ebook.c
@@ -153,22 +153,23 @@ int render_cover(scan_ebook_ctx_t *ctx, fz_context *fzctx, document_t *doc, fz_d
 
     sws_freeContext(sws_ctx);
 
-    // YUV420p -> JPEG
-    AVCodecContext *jpeg_encoder = alloc_jpeg_encoder(pixmap->w, pixmap->h, ctx->tn_qscale);
-    avcodec_send_frame(jpeg_encoder, scaled_frame);
+    // YUV420p -> JPEG/WEBP
+    AVCodecContext *thumbnail_encoder = alloc_webp_encoder(pixmap->w, pixmap->h, ctx->tn_qscale);
+    avcodec_send_frame(thumbnail_encoder, scaled_frame);
+    avcodec_send_frame(thumbnail_encoder, NULL); // Send EOF
 
-    AVPacket jpeg_packet;
-    av_init_packet(&jpeg_packet);
-    avcodec_receive_packet(jpeg_encoder, &jpeg_packet);
+    AVPacket thumbnail_packet;
+    av_init_packet(&thumbnail_packet);
+    avcodec_receive_packet(thumbnail_encoder, &thumbnail_packet);
 
     APPEND_LONG_META(doc, MetaThumbnail, 1);
-    ctx->store(doc->doc_id, 0, (char *) jpeg_packet.data, jpeg_packet.size);
+    ctx->store(doc->doc_id, 0, (char *) thumbnail_packet.data, thumbnail_packet.size);
 
     free(samples);
-    av_packet_unref(&jpeg_packet);
+    av_packet_unref(&thumbnail_packet);
     av_free(*scaled_frame->data);
     av_frame_free(&scaled_frame);
-    avcodec_free_context(&jpeg_encoder);
+    avcodec_free_context(&thumbnail_encoder);
 
     fz_drop_pixmap(fzctx, pixmap);
     fz_drop_page(fzctx, cover);

diff --git a/third-party/libscan/libscan/media/media.c b/third-party/libscan/libscan/media/media.c
@@ -68,7 +68,7 @@ void *scale_frame(const AVCodecContext *decoder, const AVFrame *frame, int size)
 
     struct SwsContext *sws_ctx = sws_getContext(
             decoder->width, decoder->height, decoder->pix_fmt,
-            dstW, dstH, AV_PIX_FMT_YUVJ420P,
+            dstW, dstH, AV_PIX_FMT_YUV420P,
             SIST_SWS_ALGO, 0, 0, 0
     );
 
@@ -436,7 +436,8 @@ int decode_frame_and_save_thumbnail(scan_media_ctx_t *ctx, AVFormatContext *pFor
         }
 
         if (seek_ok == FALSE && thumbnail_index != 0) {
-            CTX_LOG_WARNING(doc->filepath, "(media.c) Could not seek media file. Can't generate additional thumbnails.");
+            CTX_LOG_WARNING(doc->filepath,
+                            "(media.c) Could not seek media file. Can't generate additional thumbnails.");
             return SAVE_THUMBNAIL_FAILED;
         }
     }
@@ -470,34 +471,35 @@ int decode_frame_and_save_thumbnail(scan_media_ctx_t *ctx, AVFormatContext *pFor
 
         ctx->store(doc->doc_id, 0, frame_and_packet->packet->data, frame_and_packet->packet->size);
     } else {
-        // Encode frame to jpeg
-        AVCodecContext *jpeg_encoder = alloc_jpeg_encoder(scaled_frame->width, scaled_frame->height,
-                                                          ctx->tn_qscale);
-        avcodec_send_frame(jpeg_encoder, scaled_frame);
+        // Encode frame
+        AVCodecContext *thumbnail_encoder = alloc_webp_encoder(scaled_frame->width, scaled_frame->height,
+                                                               ctx->tn_qscale);
+        avcodec_send_frame(thumbnail_encoder, scaled_frame);
+        avcodec_send_frame(thumbnail_encoder, NULL); // send EOF
 
-        AVPacket jpeg_packet;
-        av_init_packet(&jpeg_packet);
-        avcodec_receive_packet(jpeg_encoder, &jpeg_packet);
+        AVPacket thumbnail_packet;
+        av_init_packet(&thumbnail_packet);
+        avcodec_receive_packet(thumbnail_encoder, &thumbnail_packet);
 
         // Save thumbnail
         if (thumbnail_index == 0) {
-            ctx->store(doc->doc_id, 0, jpeg_packet.data, jpeg_packet.size);
+            ctx->store(doc->doc_id, 0, thumbnail_packet.data, thumbnail_packet.size);
             return_value = SAVE_THUMBNAIL_OK;
 
         } else if (thumbnail_index > 1) {
             // TO FIX: the 2nd rendered frame is always broken, just skip it until
             //  I figure out a better fix.
             thumbnail_index -= 1;
 
-            ctx->store(doc->doc_id, thumbnail_index, jpeg_packet.data, jpeg_packet.size);
+            ctx->store(doc->doc_id, thumbnail_index, thumbnail_packet.data, thumbnail_packet.size);
 
             return_value = SAVE_THUMBNAIL_OK;
         } else {
             return_value = SAVE_THUMBNAIL_SKIPPED;
         }
 
-        avcodec_free_context(&jpeg_encoder);
-        av_packet_unref(&jpeg_packet);
+        avcodec_free_context(&thumbnail_encoder);
+        av_packet_unref(&thumbnail_packet);
         av_free(*scaled_frame->data);
         av_frame_free(&scaled_frame);
     }
@@ -854,9 +856,10 @@ int store_image_thumbnail(scan_media_ctx_t *ctx, void *buf, size_t buf_len, docu
         ctx->store(doc->doc_id, 0, frame_and_packet->packet->data, frame_and_packet->packet->size);
     } else {
         // Encode frame to jpeg
-        AVCodecContext *jpeg_encoder = alloc_jpeg_encoder(scaled_frame->width, scaled_frame->height,
+        AVCodecContext *jpeg_encoder = alloc_webp_encoder(scaled_frame->width, scaled_frame->height,
                                                           ctx->tn_qscale);
         avcodec_send_frame(jpeg_encoder, scaled_frame);
+        avcodec_send_frame(jpeg_encoder, NULL); // Send EOF
 
         AVPacket jpeg_packet;
         av_init_packet(&jpeg_packet);

diff --git a/third-party/libscan/libscan/media/media.h b/third-party/libscan/libscan/media/media.h
@@ -48,6 +48,28 @@ static AVCodecContext *alloc_jpeg_encoder(int w, int h, int qscale) {
     return jpeg;
 }
 
+static AVCodecContext *alloc_webp_encoder(int w, int h, int qscale) {
+
+    const AVCodec *webp_codec = avcodec_find_encoder(AV_CODEC_ID_WEBP);
+    AVCodecContext *webp = avcodec_alloc_context3(webp_codec);
+    webp->width = w;
+    webp->height = h;
+    webp->time_base.den = 1000000;
+    webp->time_base.num = 1;
+    webp->compression_level = 6;
+    webp->global_quality = FF_QP2LAMBDA * qscale;
+
+    webp->pix_fmt = AV_PIX_FMT_YUV420P;
+    webp->color_range = AVCOL_RANGE_JPEG;
+    int ret = avcodec_open2(webp, webp_codec, NULL);
+
+    if (ret != 0) {
+        return NULL;
+    }
+
+    return webp;
+}
+
 
 void parse_media(scan_media_ctx_t *ctx, vfile_t *f, document_t *doc, const char *mime_str);
 

diff --git a/third-party/libscan/libscan/raw/raw.c b/third-party/libscan/libscan/raw/raw.c
@@ -52,7 +52,7 @@ int store_thumbnail_rgb24(scan_raw_ctx_t *ctx, libraw_processed_image_t *img, do
 
     struct SwsContext *sws_ctx = sws_getContext(
             img->width, img->height, AV_PIX_FMT_RGB24,
-            dstW, dstH, AV_PIX_FMT_YUVJ420P,
+            dstW, dstH, AV_PIX_FMT_YUV420P,
             SIST_SWS_ALGO, 0, 0, 0
     );
 
@@ -76,20 +76,21 @@ int store_thumbnail_rgb24(scan_raw_ctx_t *ctx, libraw_processed_image_t *img, do
 
     sws_freeContext(sws_ctx);
 
-    AVCodecContext *jpeg_encoder = alloc_jpeg_encoder(scaled_frame->width, scaled_frame->height, 1.0f);
-    avcodec_send_frame(jpeg_encoder, scaled_frame);
+    AVCodecContext *thumbnail_encoder = alloc_webp_encoder(scaled_frame->width, scaled_frame->height, ctx->tn_qscale);
+    avcodec_send_frame(thumbnail_encoder, scaled_frame);
+    avcodec_send_frame(thumbnail_encoder, NULL); // Send EOF
 
-    AVPacket jpeg_packet;
-    av_init_packet(&jpeg_packet);
-    avcodec_receive_packet(jpeg_encoder, &jpeg_packet);
+    AVPacket thumbnail_packet;
+    av_init_packet(&thumbnail_packet);
+    avcodec_receive_packet(thumbnail_encoder, &thumbnail_packet);
 
     APPEND_LONG_META(doc, MetaThumbnail, 1);
-    ctx->store((char *) doc->doc_id, sizeof(doc->doc_id), (char *) jpeg_packet.data, jpeg_packet.size);
+    ctx->store((char *) doc->doc_id, sizeof(doc->doc_id), (char *) thumbnail_packet.data, thumbnail_packet.size);
 
-    av_packet_unref(&jpeg_packet);
+    av_packet_unref(&thumbnail_packet);
     av_free(*scaled_frame->data);
     av_frame_free(&scaled_frame);
-    avcodec_free_context(&jpeg_encoder);
+    avcodec_free_context(&thumbnail_encoder);
 
     return TRUE;
 }