#include "tesseract/ccstruct/imagedata.h" #include #include using tesseract::DocumentCache; using tesseract::DocumentData; using tesseract::ImageData; namespace { // Tests the caching mechanism of DocumentData/ImageData. class ImagedataTest : public ::testing::Test { protected: ImagedataTest() {} // Creates a fake DocumentData, writes it to a file, and returns the filename. string MakeFakeDoc(int num_pages, int doc_id, std::vector* page_texts) { // The size of the fake images that we will use. const int kImageSize = 1048576; // Not using a real image here - just an array of zeros! We are just testing // that the truth text matches. std::vector fake_image(kImageSize, 0); DocumentData write_doc("My document"); for (int p = 0; p < num_pages; ++p) { // Make some fake text that is different for each page and save it. page_texts->push_back( StringPrintf("Page %d of %d in doc %d", p, num_pages, doc_id)); // Make an imagedata and put it in the document. ImageData* imagedata = ImageData::Build("noname", p, "eng", fake_image.data(), fake_image.size(), (*page_texts)[p].c_str(), nullptr); EXPECT_EQ(kImageSize, imagedata->MemoryUsed()); write_doc.AddPageToDocument(imagedata); } // Write it to a file. string filename = file::JoinPath( FLAGS_test_tmpdir, absl::StrCat("documentdata", doc_id, ".lstmf")); EXPECT_TRUE(write_doc.SaveDocument(filename.c_str(), nullptr)); return filename; } }; TEST_F(ImagedataTest, CachesProperly) { // This test verifies that Imagedata can be stored in a DocumentData and a // collection of them is cached correctly given limited memory. // Number of pages to put in the fake document. const int kNumPages = 12; // Allowances to read the document. Big enough for 1, 3, 0, all pages. const int kMemoryAllowances[] = {2000000, 4000000, 1000000, 100000000, 0}; // Order in which to read the pages, with some sequential and some seeks. const int kPageReadOrder[] = {0, 1, 2, 3, 8, 4, 5, 6, 7, 11, 10, 9, -1}; std::vector page_texts; string filename = MakeFakeDoc(kNumPages, 0, &page_texts); // Now try getting it back with different memory allowances and check that // the pages can still be read. for (int m = 0; kMemoryAllowances[m] > 0; ++m) { DocumentData read_doc("My document"); EXPECT_TRUE( read_doc.LoadDocument(filename.c_str(), 0, kMemoryAllowances[m], nullptr)); LOG(ERROR) << "Allowance = " << kMemoryAllowances[m]; // Read the pages in a specific order. for (int p = 0; kPageReadOrder[p] >= 0; ++p) { int page = kPageReadOrder[p]; const ImageData* imagedata = read_doc.GetPage(page); EXPECT_NE(reinterpret_cast(nullptr), imagedata); // Check that this is the right page. EXPECT_STREQ(page_texts[page].c_str(), imagedata->transcription().string()); } } } TEST_F(ImagedataTest, CachesMultiDocs) { // This test verifies that DocumentCache works to store multiple DocumentData // and the two caching strategies read images in the right order. // Number of pages in each document. const std::vector kNumPages = {6, 5, 7}; std::vector> page_texts; GenericVector filenames; for (int d = 0; d < kNumPages.size(); ++d) { page_texts.emplace_back(std::vector()); string filename = MakeFakeDoc(kNumPages[d], d, &page_texts.back()); filenames.push_back(STRING(filename.c_str())); } // Now try getting them back with different cache strategies and check that // the pages come out in the right order. DocumentCache robin_cache(8000000); robin_cache.LoadDocuments(filenames, tesseract::CS_ROUND_ROBIN, nullptr); DocumentCache serial_cache(8000000); serial_cache.LoadDocuments(filenames, tesseract::CS_SEQUENTIAL, nullptr); for (int p = 0; p <= 21; ++p) { LOG(INFO) << "Page " << p; const ImageData* robin_data = robin_cache.GetPageBySerial(p); const ImageData* serial_data = serial_cache.GetPageBySerial(p); CHECK(robin_data != nullptr); CHECK(serial_data != nullptr); int robin_doc = p % kNumPages.size(); int robin_page = p / kNumPages.size() % kNumPages[robin_doc]; // Check that this is the right page. EXPECT_STREQ(page_texts[robin_doc][robin_page].c_str(), robin_data->transcription().string()); int serial_doc = p / kNumPages[0] % kNumPages.size(); int serial_page = p % kNumPages[0] % kNumPages[serial_doc]; EXPECT_STREQ(page_texts[serial_doc][serial_page].c_str(), serial_data->transcription().string()); } } } // namespace.