Jointly Optimizing Preprocessing and Inference for DNN-based Visual Analytics

Summary: Measurement study finds visual DNN analytics bottlenecked by preprocessing—not inference—on modern accelerators. SMOL exploits native low-resolution data and hardware-aware CPU/GPU co-scheduling, memory, and threading for up to 5.9× throughput at fixed accuracy. (summarized by gpt-5.6-luna on Jul 24 2026)

Paper ID: 12790
Venue: VLDB
Year: 2021
Pagerank: 7.5936939e-05
Overall Rank: 3,253 | 77.69%
DOI: 10.14778/3425879.3425881

Incoming Non-self Citations Over Time

Authors

1. Daniel Kang (Stanford University)
2. Ankit Mathur (Stanford University)
3. Teja Veeramacheneni (Stanford University)
4. Peter Bailis (Stanford University)
5. Matei Zaharia (Stanford University)

BibTeX Citation

@article{kang_vldb21,
        title = {{Jointly Optimizing Preprocessing and Inference for DNN-based Visual Analytics}},
        author = {Kang, Daniel and Mathur, Ankit and Veeramacheneni, Teja and Bailis, Peter and Zaharia, Matei},
        journal = {PVLDB},
        series = {{VLDB} '21},
        volume = {14},
        number = {2},
        pages = {87--100},
        doi = {10.14778/3425879.3425881},
        url = {https://doi.org/10.14778/3425879.3425881},
        year = {2021}
}

Incoming Citations (Sorted by Pagerank)

Showing 14 of 14 citing papers.

Rank	Citing Paper	Year	Venue	Pagerank
2,018	tf.data: A Machine Learning Data Processing Framework	2021	VLDB	9.3001937e-05
2,865	End-to-end Optimization of Machine Learning Prediction Queries	2022	SIGMOD	8.0180243e-05
3,541	Where Is My Training Bottleneck? Hidden Trade-Offs in Deep Learning Preprocessing Pipelines	2022	SIGMOD	7.3280673e-05
3,720	VSS: A Storage System for Video Analytics	2021	SIGMOD	7.1739329e-05
3,766	TASTI: Semantic Indexes for Machine Learning-based Queries over Unstructured Data	2022	SIGMOD	7.1430942e-05
3,907	Accelerating Approximate Aggregation Queries with Expensive Predicates	2021	VLDB	7.0278233e-05
4,007	Optimizing Video Analytics with Declarative Model Relationships	2023	VLDB	6.9632395e-05
4,256	VIVA: An End-to-End System for Interactive Video Analytics	2022	CIDR	6.8018439e-05
6,900	Apt-Serve: Adaptive Request Scheduling on Hybrid Cache for Scalable LLM Inference Serving	2025	SIGMOD	5.7430032e-05
7,648	Accelerating Aggregation Queries on Unstructured Streams of Data	2023	VLDB	5.575838e-05
8,589	Optimizing Video Selection LIMIT Queries With Commonsense Knowledge	2024	VLDB	5.4065627e-05
9,881	The Image Calculator: 10x Faster Image-AI Inference by Replacing JPEG with Self-designing Storage Format	2024	SIGMOD	5.2040783e-05
9,924	TVM: A Tile-based Video Management Framework	2024	VLDB	5.1955087e-05
11,162	Predictive and Near-Optimal Sampling for View Materialization in Video Databases	2024	SIGMOD	5.093636e-05

Outgoing Citations (Sorted by Pagerank)

Showing 6 of 6 cited papers.

Citations counted here include only citations to other VLDB/SIGMOD/CIDR/PODS papers in this database.

Rank	Cited Paper	Year	Venue	Pagerank
284	NoScope: Optimizing Neural Network Queries over Video at Scale	2017	VLDB	0.00022370521
295	Accelerating Machine Learning Inference with Probabilistic Predicates	2018	SIGMOD	0.00022238183
569	BlazeIt: Optimizing Declarative Aggregation and Limit Queries for Neural Network-Based Video Analytics	2020	VLDB	0.00016348191
1,229	Weld: A Common Runtime for High Performance Data Analytics	2017	CIDR	0.00011578425
1,607	Challenges and Opportunities in DNN-Based Video Analytics: A Demonstration of the BlazeIt Video Query Engine	2019	CIDR	0.0001022751
2,898	Approximate Selection with Guarantees using Proxies	2020	VLDB	7.978725e-05

Semantically Similar Papers

#	Overall Rank	Paper	Year	Venue
1	3,764	FastFlow: Accelerating Deep Learning Model Training with Smart Offloading of Input Data Pipeline	2023	VLDB
2	9,757	Faster Visual Analytics through Pixel-Perfect Aggregation	2014	VLDB
3	9,826	ETO: Accelerating Optimization of DNN Operators by High-Performance Tensor Program Reuse	2022	VLDB
4	13,432	Using Deep Learning Models to Replace Large Materialized Views in Relational Database	2021	CIDR
5	3,573	VISTA: Optimized System for Declarative Feature Transfer from Deep CNNs at Scale	2020	SIGMOD
6	2,987	Incremental and Approximate Inference for Faster Occlusion-based Deep CNN Explanations	2019	SIGMOD
7	1,607	Challenges and Opportunities in DNN-Based Video Analytics: A Demonstration of the BlazeIt Video Query Engine	2019	CIDR
8	2,384	DeepLens: Towards a Visual Data Management System	2019	CIDR
9	1,446	Analyzing and Mitigating Data Stalls in DNN Training	2021	VLDB
10	3,541	Where Is My Training Bottleneck? Hidden Trade-Offs in Deep Learning Preprocessing Pipelines	2022	SIGMOD