GeoSolver: Scaling Test-Time Reasoning in Remote Sensing with Fine-Grained Process Supervision

Dimensions overall score 8.0

• GeoPRM has remarkable cross-model generalization capabilities.result 80%
  Evidencepartial

  highlighting its remarkable cross-model generalization.

  Implicationpartial

  Strongly implied by its ability to enhance general-purpose VLMs, though not explicitly quantified

  Verificationpartial

  partial

• GeoSolver achieves state-of-the-art performance across diverse remote sensing benchmarks.result 90%
  Evidencepartial

  Extensive experiments demonstrate that our resulting model, GeoSolver-9B, achieves state-of-the-art performance across diverse remote sensing benchmarks.

  Implicationpartial

  Explicitly stated in abstract with strong assertion of results

  Verificationpartial

  partial

• GeoPRM enables robust Test-Time Scaling (TTS) that enhances both GeoSolver-9B and general-purpose VLMs.technical 85%
  Evidencepartial

  GeoPRM unlocks robust Test-Time Scaling (TTS). Serving as a universal geospatial verifier, it seamlessly scales the performance of GeoSolver-9B and directly enhances general-purpose VLMs.

  Implicationpartial

  Directly stated in abstract with clear functional description

  Verificationpartial

  partial

• Geo-PRM-2M is a large-scale, token-level process supervision dataset synthesized via entropy-guided Monte Carlo Tree Search (MCTS) and targeted visual hallucination injection.method 90%
  Evidencepartial

  We first construct Geo-PRM-2M, a large-scale, token-level process supervision dataset synthesized via entropy-guided Monte Carlo Tree Search (MCTS) and targeted visual hallucination injection.

  Implicationpartial

  Explicitly described construction method in abstract

  Verificationpartial

  partial

• Process-Aware Tree-GRPO integrates tree-structured exploration with a faithfulness-weighted reward mechanism to assign credit to intermediate steps.method 85%
  Evidencepartial

  To effectively leverage these verification signals, we propose Process-Aware Tree-GRPO, a reinforcement learning algorithm that integrates tree-structured exploration with a faithfulness-weighted reward mechanism to precisely assign credit to intermediate steps.

  Implicationpartial

  Direct technical description of the proposed algorithm

  Verificationpartial

  partial

• Ensuring visual faithfulness of intermediate reasoning steps is a critical bottleneck in remote sensing VLMs.limitation 75%
  Evidencepartial

  ensuring the visual faithfulness of these intermediate steps remains a critical bottleneck.

  Implicationpartial

  Presented as motivation for the work, supported by domain context

  Verificationpartial

  partial

• GeoSolver transitions remote sensing reasoning toward verifiable, process-supervised reinforcement learning.method 85%
  Evidencepartial

  we introduce GeoSolver, a novel framework that transitions remote sensing reasoning toward verifiable, process-supervised reinforcement learning.

  Implicationpartial

  Direct statement of the framework's contribution and approach

  Verificationpartial

  partial

• GeoPRM is a token-level process reward model that provides granular faithfulness feedback.technical 90%
  Evidencepartial

  we train GeoPRM, a token-level process reward model (PRM) that provides granular faithfulness feedback.

  Implicationpartial

  Explicit technical specification of the model's function

  Verificationpartial

  partial

Competitive landscape

GeoSolver enhances remote sensing interpretation through verifiable, process-supervised reasoning.