Pith. sign in

REVIEW 5 major objections 5 minor 1 cited by

Optimizing Domain-Specific Image Retrieval: A Benchmark of FAISS and Annoy with Fine-Tuned Features

T0 review · 5 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read On a fashion image retrieval benchmark, FAISS product quantization hits 98.4% precision with a 0.24 MB index while Annoy answers queries in about 16 microseconds.

desk verdict Plausible qualitative trade-offs, but the headline numbers rest on a results table with order-of-magnitude internal contradictions; not publishable as-is. read the letter →

arxiv 2412.01555 v1 pith:IUYZCKRK submitted 2024-12-02 cs.CV

classification cs.CV
keywords approximatenearestneighborsearchFAISSAnnoyimageretrievalResNet50fine-tuningproductquantizationfashiondatasetbenchmark
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper asks how much the approximate-nearest-neighbor (ANN) index, rather than the feature extractor, determines end-to-end image retrieval performance. It fine-tunes a ResNet50 on a 32-class fashion product dataset and feeds the same embeddings into seven FAISS index types and three Annoy configurations. The central result is a concrete trade-off: FAISS product quantization reaches 98.4% precision with a 0.24 MB index, while Annoy's angular index is the fastest at roughly 16 microseconds per query with 93.6% precision and 95.4% recall at $k=5$. A sympathetic reading is that the paper establishes a per-application selection rule: choose Annoy when latency is critical and FAISS PQ or HNSW when accuracy dominates.

What carries the argument

The machinery is a controlled end-to-end benchmark pipeline: a fine-tuned ResNet50 with its classification head removed produces one embedding per image, and the same embedding set is indexed by every ANN configuration. The named objects are the index types: FAISS Flat-L2, Flat-IP, product quantization (PQ), IVF-PQ, IVF-SQ, LSH, and HNSW, plus Annoy with angular, Euclidean, and Manhattan distances. Product quantization works by splitting each vector into subvectors and quantizing each part into a compact code; this is the mechanism that lets PQ match Flat-L2's precision while shrinking the index to 0.24 MB. Because every method receives identical inputs, the benchmark isolates the contribution of the index and the distance metric.

What would settle it

Recompute the same benchmark with ground truth defined by the original 44 category labels or by human relevance judgments; if the FAISS PQ precision lead over Annoy angular falls from its reported roughly 4.8 points to near zero, the trade-off is an artifact of the 32-class label merge or the query sample.

Watch

Extended reading notes

Core claim

On a custom 32-class fashion product dataset, after fine-tuning a ResNet50 to 98.25% validation accuracy, the paper claims that the index configuration, not the embedding alone, decides the practical trade-off among speed, accuracy, and memory. The benchmark shows FAISS Flat-L2 and FAISS product quantization both reach a precision of 0.984 at $k=5$, with PQ doing so at a 0.24 MB index; FAISS HNSW posts the highest recall (98.5%) at $k=5$, while Annoy with the angular metric reaches 93.6% precision and 95.4% recall at $k=5$ with an average query time near 16 microseconds. The claim is that no single method dominates: PQ is for memory-constrained high-accuracy pipelines, Annoy angular is for latency-critical ones, and HNSW is for recall-critical ones.

Load-bearing premise

The load-bearing premise is that the 32 hand-merged subcategory labels are exact relevance and that the 1,000 randomly sampled queries represent the retrieval task.

Editorial extensions

If this is right

  • Memory-constrained search over fashion images can use FAISS PQ to get near-exact precision at a 0.24 MB index.
  • Latency-critical retrieval can use Annoy angular, accepting roughly five points of precision for microsecond queries.
  • At $k=5$, exact and compressed FAISS indexes are effectively interchangeable, so quantization does not cost accuracy at small neighbor counts.
  • FAISS methods degrade more than Annoy as the neighbor count grows, so the choice of $k$ should be tuned together with the index family.
  • For recall-critical analytics, HNSW is the strongest configuration, at the cost of the largest index and slowest query time.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper leaves untested how much of the 4.8-point precision gap would close if Annoy's tree count were raised above the fixed value of 10.
  • Because the benchmark defines relevance by the merged 32 labels, a re-run against the original 44 categories could shrink or reverse the FAISS-over-Annoy precision lead.
  • The reported memory and index sizes are for one dataset split; on a full catalog, HNSW's graph structure should consume memory faster than PQ, so the resource ranking may not scale.
  • The low Recall@5 numbers around 0.51 suggest the top-6 voting rule underuses the recall available in the indexes, so a different voting scheme might change which method wins.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The manuscript benchmarks FAISS and Annoy for content-based image retrieval on a custom fashion dataset, after fine-tuning a ResNet50 to produce feature embeddings. It reports indexing time, memory usage, query time, precision, recall, F1-score, and Recall@5 for several FAISS index types (Flat-L2, Flat-IP, PQ, IVF-PQ, IVF-SQ, LSH, HNSW) and three Annoy distance metrics (angular, Euclidean, Manhattan). The central advertised claims are that FAISS PQ reaches 98.40% precision with a 0.24 MB index, and that Annoy is fastest with ~0.00015 s average query time. The paper concludes that Annoy angular offers the best overall balance for speed and memory constrained scenarios, while FAISS HNSW or PQ are preferable when precision is paramount.

Significance. The paper addresses a practically relevant question: how fine-tuned deep features interact with ANN index structures in an end-to-end retrieval pipeline. A careful benchmark comparing FAISS and Annoy on the same embeddings with a consistent evaluation protocol would be useful guidance for practitioners. The qualitative trade-off direction (PQ is compact and accurate, Annoy is fast) is plausible and worth stating. However, the quantitative support is not trustworthy. The central Table 2 is internally inconsistent, contradicts the prose and figures in order-of-magnitude ways, and reports physically impossible index sizes. Because the main claims are exactly the numbers in that table, and because no code, logs, or exact-search ground truth are provided, the benchmark's quantitative conclusions are currently unsupported.

major comments (5)
  1. [Table 2 (Section IV-D)] Table 2 is internally inconsistent for the FAISS rows. For example, the row labeled 'Faiss (L2)' lists Precision 0.00031565 while simultaneously listing Recall 0.98 and F1-score 0.984. Since F1 is the harmonic mean of precision and recall, it cannot exceed both inputs, and a precision of 0.0003 would imply an F1 near 0.0006, not 0.984. The same pattern appears for FAISS (PQ), (IVF-PQ), (IVF-SQ), and (HNSW). These numbers are not merely transcription slips; they invalidate the headline precision values reported in the abstract, Section IV-C, and Figure 6.
  2. [Table 2 and Section IV-C (FAISS index size and query time)] The reported index sizes and query times for FAISS Flat-L2 are physically impossible for the stated data. With 44,446 images and 2048-dimensional float32 embeddings, a flat exact index requires N*d*4 ≈ 364 MB, not the 1.17 MB shown. The same row reports an average query time of 1,671,185 µs (1.67 s) for both 'Faiss (L2)' and 'Faiss (IP)', yet Section IV-C and Figure 6 report that FAISS (L2) achieves 803.3 QPS (about 1.2 ms per query). These order-of-magnitude contradictions between the table, figure, and prose mean that the precision/memory/latency trade-off claims in the abstract are unreproducible as printed.
  3. [Section IV-D (Annoy query speed)] The prose in Section IV-D states that 'Annoy angular achieved 53 queries per second' while the same section later reports an average query time of 16 µs. These two numbers are mutually inconsistent: 16 µs per query corresponds to roughly 62,500 QPS, not 53 QPS. If the 53 QPS figure is correct, the query time should be about 18,868 µs, not 16 µs. Because the paper's main speed claim for Annoy is 0.00015 s in the abstract, this contradiction directly affects a central result.
  4. [Evaluation Protocol (Section III-D); Results (Section IV)] No reproducibility artifacts are provided: there is no code, no raw result logs, no random seed, no error bars or standard deviations across multiple runs, and no comparison against an exact nearest-neighbor baseline. Given that the reported numbers are not self-consistent and appear to mix different measurement units (e.g., memory values near 0.0001 MB for a 44k-vector index), the reader cannot determine which entries are correct or reconstruct the actual benchmark. This is a load-bearing gap because the paper's conclusions are exclusively quantitative.
  5. [Dataset preprocessing (Section III-A) and evaluation protocol (Section III-D)] The evaluation uses the 32 merged/split subcategory labels as exact relevance ground truth and a single random sample of 1,000 queries, with no sensitivity analysis. While this is a common practical choice, the merging of visually similar categories (e.g., Scarves/Mufflers, Topwear/Shirts) and the absence of any label-noise or query-sampling robustness check means that every reported precision/recall/F1 value could shift if the label definitions or query set change. This concern is secondary to the internal numeric inconsistencies, but it should be addressed if the benchmark is to support generalizable guidance.
minor comments (5)
  1. [Section III-B] The formula labeled 'Recall' is mislabeled as 'Precision = TP/(TP+FN)' in the rendered text; it should read 'Recall = TP/(TP+FN)'.
  2. [Throughout the manuscript] The acronym 'ANN' is used for both 'Artificial Neural Network' (Introduction) and 'Approximate Nearest Neighbor' (throughout the rest of the paper). This is confusing in a paper about ANN search with a fine-tuned neural network; consider using 'ANN search' or 'ANNS' consistently.
  3. [References] Reference [18] duplicates [12] (ParlayANN), and [37] duplicates [17] (Malkov and Yashunin). Please deduplicate.
  4. [Figures 3, 7, and Table 1] The paper states the model was fine-tuned over 10 epochs in Figure 1 and in the text, but Table 1 lists 11 epochs; also the caption says '10 epochs' while the table has 11 rows. Please make the epoch counts consistent and check corresponding validation metrics.
  5. [Section IV-B and Table 2] The prose in Section IV-B says Angular achieves 'a precision of 0.9443' at 5 neighbors, but Table 2 lists Annoy (Angular) precision as 0.9357. Similarly, the text in Section IV-D reports 93.56% precision for Annoy angular while Table 2 gives 0.93565575; either round consistently or clarify the discrepancy.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the benchmark is empirical and self-contained, with no fitted parameter or self-citation chain doing load-bearing work.

full rationale

This paper is an empirical benchmark, not a derivation. The pipeline is: fine-tune ResNet50 on a fashion dataset, extract embeddings, index them with FAISS and Annoy variants, run held-out queries, and report precision/recall/F1/query-time/memory. None of the paper's equations (Adam updates, learning-rate decay) is used to define the reported retrieval metrics, and no fitted parameter is renamed as a prediction. The same subcategory labels are used for training and for computing retrieval relevance, but the test set is held out and precision is computed from retrieved neighbors against those test labels, so the evaluation is not forced by construction. The only self-citation is reference [23], a prior related paper by overlapping authors; it is cited in the background section as an example of Faiss's flexibility and is not load-bearing for any claim in this paper. The headline numbers from Table 2 may be internally inconsistent or physically implausible—e.g., FAISS L2 index size of 1.17 MB for 44k 2048-dimensional float32 vectors is far too small, and several precision values in the table contradict the prose—but those are reproducibility/correctness concerns, not circularity. There is no step where an output metric reduces to an input by definition or where a conclusion is justified solely by an unverified self-citation. Accordingly, the circularity score is 0.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The central benchmark rests on hand-set ANN hyperparameters and on the assumption that class labels define retrieval relevance. No code or data artifacts are provided, so the reported numbers cannot be independently reconstructed.

free parameters (5)
  • Annoy n_trees = 10
    Set in Section III-C; directly controls the accuracy and speed trade-off of all Annoy results.
  • HNSW M = 32
    Graph connectivity set in Section III-C for FAISS HNSW; affects recall and index size.
  • HNSW efConstruction = 40
    Recall quality parameter set in Section III-C; affects index construction quality and time.
  • top_k retrieved neighbors = 6 (query excluded, effective k=5)
    Evaluation protocol in Section III-D retrieves top 6 neighbors excluding the query; all precision, recall, F1, and Recall@5 values depend on this choice.
  • Initial learning rate and scheduler factor = 0.001 and 0.1
    Training hyperparameters in Section III-B determine the embedding quality that all ANN results depend on.
assumptions (3)
  • domain assumption Subcategory labels of the custom 32-class Fashion subset are a valid relevance ground truth for image retrieval.
    Used to compute precision, recall, F1, and Recall@5; any label noise from category merging in Section III-A changes all reported metrics.
  • domain assumption The 1,000-query random test subset is representative of the test distribution.
    Section III-D selects queries randomly but reports no seed or stratification; a biased sample would skew comparative results.
  • domain assumption Fine-tuned ResNet50 embeddings capture the visual similarity needed for nearest-neighbor retrieval.
    Section III-B assumes classification-trained features transfer to retrieval; no pretrained-feature baseline is provided to test this.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Optimizing Domain-Specific Image Retrieval: A Benchmark of FAISS and Annoy with Fine-Tuned Features." pith.science (2026). https://pith.science/paper/IUYZCKRK

@misc{pith2026241201555,
  author       = {Pith},
  title        = {Pith review of: Optimizing Domain-Specific Image Retrieval: A Benchmark of FAISS and Annoy with Fine-Tuned Features},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IUYZCKRK}},
  note         = {Machine review of arXiv:2412.01555}
}
read the original abstract

Approximate Nearest Neighbor search is one of the keys to high-scale data retrieval performance in many applications. The work is a bridge between feature extraction and ANN indexing through fine-tuning a ResNet50 model with various ANN methods: FAISS and Annoy. We evaluate the systems with respect to indexing time, memory usage, query time, precision, recall, F1-score, and Recall@5 on a custom image dataset. FAISS's Product Quantization can achieve a precision of 98.40% with low memory usage at 0.24 MB index size, and Annoy is the fastest, with average query times of 0.00015 seconds, at a slight cost to accuracy. These results reveal trade-offs among speed, accuracy, and memory efficiency and offer actionable insights into the optimization of feature-based image retrieval systems. This study will serve as a blueprint for constructing actual retrieval pipelines and be built on fine-tuned deep learning networks and associated ANN methods.

Figures

Figures reproduced from arXiv: 2412.01555 by the authors.

Figure 2
Figure 2. Queries per second versus precision for Annoy with different distance metrics: angular, Euclidean, and Manhattan. evaluation avoided variability that might arise from different input distributions, ensuring direct comparability of the results across different indexing methods. The embeddings for all images in the test dataset were extracted using the fine-tuned ResNet50 model. These embeddings served as inputs to th… view at source ↗
Figure 3
Figure 3. presents the impact of varying the number of neighbors (k) on precision, recall, and F1-score for the three metrics. Angular consistently outperforms Euclidean and Manhattan in accuracy metrics as k increases. For instance, with 50 neighbors, Angular achieves an F1-score of 0.8825, while Euclidean and Manhattan achieve 0.8743 and 0.8724, TABLE 1: Training and Validation metrics for each epoch Epoch Train Loss Valida… view at source ↗
Figure 5
Figure 5. Retrieval results for fashion product queries using different distance metrics: (a) Angular, (b) Euclidean, and (c) Manhattan, demonstrating variations in nearest neighbor selection across categories [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figures from the paper (4 more)
Figure 7
Figure 7. Figure 7: Comparison of FAISS index types: (a) F1-score versus the number of neighbors, (b) precision versus the number of neighbors, and (c) queries per second versus precision, highlighting trade-offs in retrieval accuracy, speed, and neighborhood size for different indexing m…
Figure 8
Figure 8. Figure 8: Comparison of index size and memory usage (in MB) for different FAISS index types, highlighting the trade-offs between storage requirements and runtime memory efficiency. 0.26 MB for 5 neighbors, making it a strong candidate for applications requiring efficiency [PITH…
Figure 9
Figure 9. Figure 9: Retrieval results for various indexing methods: (a), (b) Flat Index (L2), (c) Flat Index (Inner Product), (d) Inverted File with IVF, (e) Product Quantization (PQ), (f) IVF with Scalar Quantization, (g) Locality-Sensitive Hashing (LSH), (h) HNSW [PITH_FULL_IMAGE:figur…
Figure 10
Figure 10. Figure 10: Comparative analysis of performance metrics across indexing methods: (a) Queries Per Second vs. Precision, (b) F1- Score vs. Neighbors, (c) Recall vs. Neighbors, and (d) Memory Usage and Index Size vs. Metric [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. GPU-Accelerated ANNS: Quantized for Speed, Built for Change

    cs.DB 2026-01 conditional novelty 6.0 of 10

    Jasper, a GPU-native Vamana ANNS system, combines batch-parallel construction, RaBitQ quantization, and an optimized beam-search kernel to deliver reported throughput up to ~1.9x CAGRA with streaming updates.

Reference graph

Works this paper leans on

37 extracted references · 21 canonical work pages · cited by 1 Pith paper

  1. [1]

    AI revolutionizing industries worldwide: A comprehensive overview of its diverse applications,

    A. Bin Rashid and M. A. K. Kausik, “AI revolutionizing industries worldwide: A comprehensive overview of its diverse applications,” Hybrid Advances, vol. 7, p. 100277, Dec. 2024, doi: 10.1016/J.HYBADV.2024.100277

  2. [2]

    E- Commerce Recommendation Applications,

    J. Ben Schafer, J. A. Konstan, and J. Riedl, “E- Commerce Recommendation Applications,” Data Mining and Knowledge Discovery 2001 5:1, vol. 5, no. 1, pp. 115–153, 2001, doi: 10.1023/A:1009804230409

  3. [3]

    Content-based image retrieval for medical diagnosis using fuzzy clustering and deep learning,

    D. K. Sudhish, L. R. Nair, and S. S, “Content-based image retrieval for medical diagnosis using fuzzy clustering and deep learning,” Biomed Signal Process Control, vol. 88, p. 105620, Feb. 2024, doi: 10.1016/J.BSPC.2023.105620

  4. [4]

    State-of-the-art in artificial neural network applications: A survey,

    O. I. Abiodun, A. Jantan, A. E. Omolara, K. V. Dada, N. A. E. Mohamed, and H. Arshad, “State-of-the-art in artificial neural network applications: A survey,” Heliyon, vol. 4, no. 11, p. e00938, Nov. 2018, doi: 10.1016/J.HELIYON.2018.E00938

  5. [5]

    Artificial neural networks: fundamentals, computing, design, and application,

    I. A. Basheer and M. Hajmeer, “Artificial neural networks: fundamentals, computing, design, and application,” J Microbiol Methods, vol. 43, no. 1, pp. 3–31, Dec. 2000, doi: 10.1016/S0167-7012(00)00201- 3

  6. [6]

    A comprehensive review for industrial applicability of artificial neural networks,

    M. R. G. Meireles, P. E. M. Almeida, and M. G. Simões, “A comprehensive review for industrial applicability of artificial neural networks,” IEEE Transactions on Industrial Electronics, vol. 50, no. 3, pp. 585–601, Jun. 2003, doi: 10.1109/TIE.2003.812470

  7. [7]

    The Faiss library,

    M. Douze et al., “The Faiss library,” Jan. 2024, Accessed: Dec. 02, 2024. [Online]. Available: https://arxiv.org/abs/2401.08281v2

  8. [8]

    GitHub - spotify/annoy: Approximate Nearest Neighbors in C++/Python optimized for memory usage and loading/saving to disk

    “GitHub - spotify/annoy: Approximate Nearest Neighbors in C++/Python optimized for memory usage and loading/saving to disk.” Accessed: Dec. 02,

Show all 37 references
  1. [9]

    Comparison of different ANN techniques in river flow prediction,

    O. Kisi and H. K. Cigizoglu, “Comparison of different ANN techniques in river flow prediction,” Civil Engineering and Environmental Systems, vol. 24, no. 3, pp. 211–231, Sep. 2007, doi: 10.1080/10286600600888565

  2. [10]

    Domain-specific language models pre-trained on construction management systems corpora,

    Y. Zhong and S. D. Goodfellow, “Domain-specific language models pre-trained on construction management systems corpora,” Autom Constr, vol. 160, p. 105316, Apr. 2024, doi: 10.1016/J.AUTCON.2024.105316

  3. [11]

    Improving Approximate Nearest Neighbor Search through Learned Adaptive Early Termination,

    C. Li, M. Zhang, D. G. Andersen, and Y. He, “Improving Approximate Nearest Neighbor Search through Learned Adaptive Early Termination,” Proceedings of the ACM SIGMOD International Conference on Management of Data, pp. 2539–2554, Jun. 2020, doi: 10.1145/3318464.3380600/SUPPL_FI...

  4. [13]

    Product quantization for nearest neighbor search,

    H. Jégou, M. Douze, and C. Schmid, “Product quantization for nearest neighbor search,” IEEE Trans Pattern Anal Mach Intell, vol. 33, no. 1, pp. 117–128, 2011, doi: 10.1109/TPAMI.2010.57

  5. [14]

    Billion-Scale Similarity Search with GPUs,

    J. Johnson, M. Douze, and H. Jegou, “Billion-Scale Similarity Search with GPUs,” IEEE Trans Big Data, vol. 7, no. 3, pp. 535–547, Jul. 2021, doi: 10.1109/TBDATA.2019.2921572

  6. [15]

    Curator: Efficient Indexing for Multi-Tenant Vector Databases,

    Y. Jin, Y. Wu, W. Hu, B. M. Maggs, X. Zhang, and D. Zhuo, “Curator: Efficient Indexing for Multi-Tenant Vector Databases,” Jan. 2024, Accessed: Dec. 02,

  7. [16]

    Constrained Approximate Similarity Search on Proximity Graph,

    W. Zhao, S. Tan, and P. Li, “Constrained Approximate Similarity Search on Proximity Graph,” Oct. 2022, Accessed: Dec. 02, 2024. [Online]. Available: https://arxiv.org/abs/2210.14958v2

  8. [17]

    Available: http://arxiv.org/abs/2401.07119

    [Online]. Available: http://arxiv.org/abs/2401.07119

  9. [18]

    ParlayANN: Scalable and Deterministic Parallel Graph-Based Approximate Nearest Neighbor Search Algorithms,

    M. D. Manohar et al., “ParlayANN: Scalable and Deterministic Parallel Graph-Based Approximate Nearest Neighbor Search Algorithms,” May 2023, Accessed: Dec. 02, 2024. [Online]. Available: https://arxiv.org/abs/2305.04359v2

  10. [20]

    ANN-Benchmarks: A benchmarking tool for approximate nearest neighbor algorithms,

    M. Aumüller, E. Bernhardsson, and A. Faithfull, “ANN-Benchmarks: A benchmarking tool for approximate nearest neighbor algorithms,” Inf Syst, vol. 87, p. 101374, Jan. 2020, doi: 10.1016/J.IS.2019.02.006

  11. [21]

    A meta-learning configuration framework for graph-based similarity search indexes,

    R. S. Oyamada, L. C. Shimomura, S. Barbon, and D. S. Kaster, “A meta-learning configuration framework for graph-based similarity search indexes,” Inf Syst, vol. 112, p. 102123, Feb. 2023, doi: 10.1016/J.IS.2022.102123

  12. [22]

    An efficient faiss-based search method for mass spectral library searching,

    C. Qin, C. Deng, J. Huang, K. Shu, and M. Bai, “An efficient faiss-based search method for mass spectral library searching,” Proceedings - 2020 3rd International Conference on Advanced Electronic Materials, Computers and Software Engineering, AEMCSE 2020, pp. 513–518, Apr. 202...

  13. [23]

    Approximate Similarity Search with FAISS Framework Using FPGAs on the Cloud,

    D. Danopoulos, C. Kachris, and D. Soudris, “Approximate Similarity Search with FAISS Framework Using FPGAs on the Cloud,” Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11733 LNCS, pp. ...

  14. [24]

    An Investigation of Practical Approximate Nearest Neighbor Algorithms,

    T. Liu, A. Moore, K. Yang, and A. Gray, “An Investigation of Practical Approximate Nearest Neighbor Algorithms,” Adv Neural Inf Process Syst, vol. 17, 2004

  15. [25]

    Efficient Medical Image Retrieval Using DenseNet and FAISS for BIRADS Classification,

    M. S. Rahman, F. Humayara, S. M. E. Rabbi, and M. M. Rashid, “Efficient Medical Image Retrieval Using DenseNet and FAISS for BIRADS Classification,” 14 Nov. 2024, Accessed: Dec. 02, 2024. [Online]. Available: https://arxiv.org/abs/2411.01473v1

  16. [26]

    Fast Open Modification Spectral Library Searching through Approximate Nearest Neighbor Indexing,

    W. Bittremieux, P. Meysman, W. S. Noble, and K. Laukens, “Fast Open Modification Spectral Library Searching through Approximate Nearest Neighbor Indexing,” J Proteome Res, vol. 17, no. 10, pp. 3463– 3474, Oct. 2018, doi: 10.1021/ACS.JPROTEOME.8B00359/SUPPL_FILE/ PR8B00359_SI_001.PDF

  17. [27]

    Survey of vector database management systems,

    J. J. Pan, J. Wang, and G. Li, “Survey of vector database management systems,” VLDB Journal, vol. 33, no. 5, pp. 1591–1615, Sep. 2024, doi: 10.1007/S00778-024-00864-X/METRICS

  18. [28]

    Fashion Product Images Dataset

    A. Param, “Fashion Product Images Dataset.” Accessed: Dec. 02, 2024. [Online]. Available: https://www.kaggle.com/datasets/paramaggarwal/fashi on-product-images-dataset

  19. [29]

    The role of local dimensionality measures in benchmarking nearest neighbor search,

    M. Aumüller and M. Ceccarello, “The role of local dimensionality measures in benchmarking nearest neighbor search,” Inf Syst, vol. 101, p. 101807, Nov. 2021, doi: 10.1016/J.IS.2021.101807

  20. [30]

    Advanced data mining techniques,

    D. L. Olson and D. Delen, “Advanced data mining techniques,” Advanced Data Mining Techniques, pp. 1–180, 2008, doi: 10.1007/978-3-540-76917- 0/COVER

  21. [31]

    Adafactor: Adaptive Learning Rates with Sublinear Memory Cost,

    N. Shazeer and M. Stern, “Adafactor: Adaptive Learning Rates with Sublinear Memory Cost,” Jul. 03, 2018, PMLR. Accessed: Dec. 02, 2024. [Online]. Available: https://proceedings.mlr.press/v80/shazeer18a.html

  22. [32]

    A Review of the F-Measure: Its History, Properties, Criticism, and Alternatives,

    P. Christen, D. J. Hand, and N. Kirielle, “A Review of the F-Measure: Its History, Properties, Criticism, and Alternatives,” ACM Comput Surv, vol. 56, no. 3, Mar. 2023, doi: 10.1145/3606367/ASSET/91B75E27- EDA5-45BA-AFA6- D20864C7414A/ASSETS/GRAPHIC/CSUR-2022- 0380-F08.JPG

  23. [33]

    Evaluation: from precision, recall and F-measure to ROC, informedness, markedness and correlation,

    D. M. W. Powers, “Evaluation: from precision, recall and F-measure to ROC, informedness, markedness and correlation,” Oct. 2020, Accessed: Dec. 02, 2024. [Online]. Available: https://arxiv.org/abs/2010.16061v1

  24. [34]

    High-dimensional signature compression for large-scale image classification,

    J. Sánchez and F. Perronnin, “High-dimensional signature compression for large-scale image classification,” Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, pp. 1665–1672, 2011, doi: 10.1109/CVPR.2011.5995504

  25. [35]

    The advantages of the Matthews correlation coefficient (MCC) over F1 score and accuracy in binary classification evaluation,

    D. Chicco and G. Jurman, “The advantages of the Matthews correlation coefficient (MCC) over F1 score and accuracy in binary classification evaluation,” BMC Genomics, vol. 21, no. 1, pp. 1–13, Jan. 2020, doi: 10.1186/S12864-019-6413-7/TABLES/5

  26. [36]

    Optimization of Indexing Based on k-Nearest Neighbor Graph for Proximity Search in High-dimensional Data,

    M. Iwasaki and D. Miyazaki, “Optimization of Indexing Based on k-Nearest Neighbor Graph for Proximity Search in High-dimensional Data,” Oct. 2018, Accessed: Dec. 02, 2024. [Online]. Available: http://arxiv.org/abs/1810.07355

  27. [37]

    Announcing ScaNN: Efficient Vector Similarity Search

    “Announcing ScaNN: Efficient Vector Similarity Search.” Accessed: Dec. 02, 2024. [Online]. Available: https://research.google/blog/announcing- scann-efficient-vector-similarity-search/

  28. [39]

    Efficient and Robust Approximate Nearest Neighbor Search Using Hierarchical Navigable Small World Graphs,

    Y. A. Malkov and D. A. Yashunin, “Efficient and Robust Approximate Nearest Neighbor Search Using Hierarchical Navigable Small World Graphs,” IEEE Trans Pattern Anal Mach Intell, vol. 42, no. 4, pp. 824–836, Apr. 2020, doi: 10.1109/TPAMI.2018.2889473

  29. [2024]

    Available: https://github.com/spotify/annoy

    [Online]. Available: https://github.com/spotify/annoy

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.