Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
EMPLACE: Self-Supervised Urban Scene Change Detection
Authors: Tim Alpherts, Sennay Ghebreab, Nanne van Noord
AAAI 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this paper we introduce AC-1M the largest USCD dataset by far of over 1.1M images, together with EMPLACE, a selfsupervising method to train a Vision Transformer using our adaptive triplet loss. We show EMPLACE outperforms SOTA methods both as a pre-training method for linear fine-tuning as well as a zero-shot setting. Lastly, in a case study of Amsterdam, we show that we are able to detect both small and large changes throughout the city and that changes uncovered by EMPLACE, depending on size, correlate with housing prices which in turn is indicative of inequity. Experiments and Evaluation In this section we will describe our procedure for training on the AC-1M, the parameters, and our evaluation procedure through order prediction. |
| Researcher Affiliation | Academia | University of Amsterdam EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes the method for triplet mining, adaptive triplet loss, and data augmentation in detail. However, it does not include a clearly labeled pseudocode block or algorithm figure. |
| Open Source Code | Yes | The details on how to retrieve the AC-1M will be available at github.com/Timalph/EMPLACE. Both datasets will be available at github.com/Timalph/EMPLACE. |
| Open Datasets | Yes | We build the AC-1M, the largest Urban Scene Change Detection dataset to date containing over 1.1M panoramas curated to be within a single meter of each other. The details on how to retrieve the AC-1M will be available at github.com/Timalph/EMPLACE. Both datasets will be available at github.com/Timalph/EMPLACE. Both the polygons and the panoramas are available from the Amsterdam Municipality API. |
| Dataset Splits | Yes | We randomly split the AC-1M dataset into a training, validation, and test set using a 70/20/10 split. This split is performed by cluster, which means every cluster only appears in one set. We split both datasets into train, validation, and test sets following a 70/20/10 split and fine-tune the EMPLACE model that scores best on order prediction to perform discrete change prediction. |
| Hardware Specification | Yes | Training and evaluation was conducted on 4 NVIDIA GeForce GTX 1080 Ti GPUs. Training and evaluation was conducted on 1 NVIDIA GeForce GTX 1080 Ti GPU. |
| Software Dependencies | No | The paper mentions using the Adam optimizer but does not specify any software libraries or their version numbers (e.g., PyTorch 1.x, TensorFlow 2.x, Python 3.x). |
| Experiment Setup | Yes | We use the Adam optimizer with a learning rate set to 10^-5, a batch size of 64, and grad norm set to 0.5. We use the Adam optimizer with a learning rate set to 10^-5, a batch size of 16, grad norm set to 0.5, and early stopping after not having improved for 3 epochs. |