{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:UI644VF22ISURCEMQ6H7LVJJU7","short_pith_number":"pith:UI644VF2","canonical_record":{"source":{"id":"2604.08213","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-09T13:11:33Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"c82fb40e08059541fad887d45a11a27ee4e2310312c3b7193b08b4a06ec074a7","abstract_canon_sha256":"3a2e4252e971168d8233c682f68285742af2df9e05ead7a570e68d4258b92e57"},"schema_version":"1.0"},"canonical_sha256":"a23dce54bad22548888c878ff5d529a7eeabbc9527d7c8edc480a0a834b81b33","source":{"kind":"arxiv","id":"2604.08213","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.08213","created_at":"2026-05-26T02:05:09Z"},{"alias_kind":"arxiv_version","alias_value":"2604.08213v2","created_at":"2026-05-26T02:05:09Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.08213","created_at":"2026-05-26T02:05:09Z"},{"alias_kind":"pith_short_12","alias_value":"UI644VF22ISU","created_at":"2026-05-26T02:05:09Z"},{"alias_kind":"pith_short_16","alias_value":"UI644VF22ISURCEM","created_at":"2026-05-26T02:05:09Z"},{"alias_kind":"pith_short_8","alias_value":"UI644VF2","created_at":"2026-05-26T02:05:09Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:UI644VF22ISURCEMQ6H7LVJJU7","target":"record","payload":{"canonical_record":{"source":{"id":"2604.08213","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-09T13:11:33Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"c82fb40e08059541fad887d45a11a27ee4e2310312c3b7193b08b4a06ec074a7","abstract_canon_sha256":"3a2e4252e971168d8233c682f68285742af2df9e05ead7a570e68d4258b92e57"},"schema_version":"1.0"},"canonical_sha256":"a23dce54bad22548888c878ff5d529a7eeabbc9527d7c8edc480a0a834b81b33","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-26T02:05:09.119027Z","signature_b64":"GjNc08To4qqRYawfAPK9hH/KWKgZ0tKkfvkg/bnzlDJfMLvLsrS8FdxMKL6+MlPkkAzmUDcHu4H21sMSIGslCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a23dce54bad22548888c878ff5d529a7eeabbc9527d7c8edc480a0a834b81b33","last_reissued_at":"2026-05-26T02:05:09.118007Z","signature_status":"signed_v1","first_computed_at":"2026-05-26T02:05:09.118007Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2604.08213","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-26T02:05:09Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"odO2KYslmUKledCGPUfWIEs3T7CsYztIkQS3HYhT/zxznqORh/bQ8ZikWkZxGRFvjwPeEPXm2e8ALjffaNggDQ==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-27T04:50:02.752781Z"},"content_sha256":"7f5f23df8719843b22f1363eff7755b843ae1fd27f5e65644b89ec63d9c8bde4","schema_version":"1.0","event_id":"sha256:7f5f23df8719843b22f1363eff7755b843ae1fd27f5e65644b89ec63d9c8bde4"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:UI644VF22ISURCEMQ6H7LVJJU7","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"EditCaption: Human-Refined SFT and HAE-DPO for Image Editing Instruction Synthesis","license":"http://creativecommons.org/licenses/by/4.0/","headline":"A two-stage SFT and DPO pipeline aligns vision-language models to cut critical errors in image editing instructions from 47% to 23%.","cross_cats":["cs.AI"],"primary_cat":"cs.CV","authors_text":"Chao Hui, Haohua Chen, Hao Shi, Honghao Cai, Tianze Zhou, Wei Zhu, Xiangyuan Wang, Xu Tang, Yao Hu, Yibo Chen, Yuling Wu, Yunhao Bai","submitted_at":"2026-04-09T13:11:33Z","abstract_excerpt":"High-quality source-target image pairs with precise editing instructions are essential for instruction-guided image editing, yet constructing such training triplets at scale remains costly. Recent pipelines often rely on vision-language models to synthesize editing instructions automatically, but we find that strong VLMs still struggle to describe visual transformations between image pairs. In particular, they exhibit three recurring failure modes: orientation inconsistency, viewpoint ambiguity, and missing fine-grained attributes. In a human evaluation on 400 image pairs, several open-source "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"fine-tuned Qwen3-VL models outperform open-source baselines; the 235B model reaches 4.712 on Eval-400 (vs. Gemini-3-Pro 4.706, GPT-4.1 4.220, Kimi-K2.5 4.111) and 4.588 on ByteMorph-Bench (vs. Gemini-3-Pro 4.522, GPT-4.1 3.412). Human evaluation shows critical errors falling from 47.75% to 23% and correctness rising from 41.75% to 66%.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the three identified failure modes (orientation inconsistency, viewpoint ambiguity, insufficient fine-grained attribute description) are the dominant sources of unusable instructions and that the human preference data collected for DPO faithfully captures them without introducing new selection biases or annotation artifacts.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"EditCaption reduces critical errors in automated image editing instructions from 47.75% to 23% via SFT and DPO, yielding fine-tuned models that match or exceed closed-source VLMs on Eval-400 and ByteMorph-Bench.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"A two-stage SFT and DPO pipeline aligns vision-language models to cut critical errors in image editing instructions from 47% to 23%.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"909b0db20a362c370bf03b8ee5d6b67936c8b3fd405faabb8c7cda36d1df6e95"},"source":{"id":"2604.08213","kind":"arxiv","version":2},"verdict":{"id":"1457be13-faa7-4e28-aa20-6eb82131bf13","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T17:18:58.364884Z","strongest_claim":"fine-tuned Qwen3-VL models outperform open-source baselines; the 235B model reaches 4.712 on Eval-400 (vs. Gemini-3-Pro 4.706, GPT-4.1 4.220, Kimi-K2.5 4.111) and 4.588 on ByteMorph-Bench (vs. Gemini-3-Pro 4.522, GPT-4.1 3.412). Human evaluation shows critical errors falling from 47.75% to 23% and correctness rising from 41.75% to 66%.","one_line_summary":"EditCaption reduces critical errors in automated image editing instructions from 47.75% to 23% via SFT and DPO, yielding fine-tuned models that match or exceed closed-source VLMs on Eval-400 and ByteMorph-Bench.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the three identified failure modes (orientation inconsistency, viewpoint ambiguity, insufficient fine-grained attribute description) are the dominant sources of unusable instructions and that the human preference data collected for DPO faithfully captures them without introducing new selection biases or annotation artifacts.","pith_extraction_headline":"A two-stage SFT and DPO pipeline aligns vision-language models to cut critical errors in image editing instructions from 47% to 23%."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.08213/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"221fb1646a566e1c3b87cb96bcb394a36edf1786dc2eeef6261669106aacea27"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"1457be13-faa7-4e28-aa20-6eb82131bf13"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-26T02:05:09Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"Yw61wGcy4yDki83jqgBhs/abCOiyDGP1oPn9AbRte1ndYx8SVBN3PSV5rseD8reQ702EbmoNB1omwYvXjel+Dg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-05-27T04:50:02.753269Z"},"content_sha256":"8490e0e17e3818c24cf1f6614e2baf3346f2b115e0a637a2e17db26707643cb4","schema_version":"1.0","event_id":"sha256:8490e0e17e3818c24cf1f6614e2baf3346f2b115e0a637a2e17db26707643cb4"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/UI644VF22ISURCEMQ6H7LVJJU7/bundle.json","state_url":"https://pith.science/pith/UI644VF22ISURCEMQ6H7LVJJU7/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/UI644VF22ISURCEMQ6H7LVJJU7/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-05-27T04:50:02Z","links":{"resolver":"https://pith.science/pith/UI644VF22ISURCEMQ6H7LVJJU7","bundle":"https://pith.science/pith/UI644VF22ISURCEMQ6H7LVJJU7/bundle.json","state":"https://pith.science/pith/UI644VF22ISURCEMQ6H7LVJJU7/state.json","well_known_bundle":"https://pith.science/.well-known/pith/UI644VF22ISURCEMQ6H7LVJJU7/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:UI644VF22ISURCEMQ6H7LVJJU7","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"3a2e4252e971168d8233c682f68285742af2df9e05ead7a570e68d4258b92e57","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-09T13:11:33Z","title_canon_sha256":"c82fb40e08059541fad887d45a11a27ee4e2310312c3b7193b08b4a06ec074a7"},"schema_version":"1.0","source":{"id":"2604.08213","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2604.08213","created_at":"2026-05-26T02:05:09Z"},{"alias_kind":"arxiv_version","alias_value":"2604.08213v2","created_at":"2026-05-26T02:05:09Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2604.08213","created_at":"2026-05-26T02:05:09Z"},{"alias_kind":"pith_short_12","alias_value":"UI644VF22ISU","created_at":"2026-05-26T02:05:09Z"},{"alias_kind":"pith_short_16","alias_value":"UI644VF22ISURCEM","created_at":"2026-05-26T02:05:09Z"},{"alias_kind":"pith_short_8","alias_value":"UI644VF2","created_at":"2026-05-26T02:05:09Z"}],"graph_snapshots":[{"event_id":"sha256:8490e0e17e3818c24cf1f6614e2baf3346f2b115e0a637a2e17db26707643cb4","target":"graph","created_at":"2026-05-26T02:05:09Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"fine-tuned Qwen3-VL models outperform open-source baselines; the 235B model reaches 4.712 on Eval-400 (vs. Gemini-3-Pro 4.706, GPT-4.1 4.220, Kimi-K2.5 4.111) and 4.588 on ByteMorph-Bench (vs. Gemini-3-Pro 4.522, GPT-4.1 3.412). Human evaluation shows critical errors falling from 47.75% to 23% and correctness rising from 41.75% to 66%."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the three identified failure modes (orientation inconsistency, viewpoint ambiguity, insufficient fine-grained attribute description) are the dominant sources of unusable instructions and that the human preference data collected for DPO faithfully captures them without introducing new selection biases or annotation artifacts."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"EditCaption reduces critical errors in automated image editing instructions from 47.75% to 23% via SFT and DPO, yielding fine-tuned models that match or exceed closed-source VLMs on Eval-400 and ByteMorph-Bench."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"A two-stage SFT and DPO pipeline aligns vision-language models to cut critical errors in image editing instructions from 47% to 23%."}],"snapshot_sha256":"909b0db20a362c370bf03b8ee5d6b67936c8b3fd405faabb8c7cda36d1df6e95"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"221fb1646a566e1c3b87cb96bcb394a36edf1786dc2eeef6261669106aacea27"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/2604.08213/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"High-quality source-target image pairs with precise editing instructions are essential for instruction-guided image editing, yet constructing such training triplets at scale remains costly. Recent pipelines often rely on vision-language models to synthesize editing instructions automatically, but we find that strong VLMs still struggle to describe visual transformations between image pairs. In particular, they exhibit three recurring failure modes: orientation inconsistency, viewpoint ambiguity, and missing fine-grained attributes. In a human evaluation on 400 image pairs, several open-source ","authors_text":"Chao Hui, Haohua Chen, Hao Shi, Honghao Cai, Tianze Zhou, Wei Zhu, Xiangyuan Wang, Xu Tang, Yao Hu, Yibo Chen, Yuling Wu, Yunhao Bai","cross_cats":["cs.AI"],"headline":"A two-stage SFT and DPO pipeline aligns vision-language models to cut critical errors in image editing instructions from 47% to 23%.","license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-09T13:11:33Z","title":"EditCaption: Human-Refined SFT and HAE-DPO for Image Editing Instruction Synthesis"},"references":{"count":0,"internal_anchors":0,"resolved_work":0,"sample":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2604.08213","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-10T17:18:58.364884Z","id":"1457be13-faa7-4e28-aa20-6eb82131bf13","model_set":{"reader":"grok-4.3"},"one_line_summary":"EditCaption reduces critical errors in automated image editing instructions from 47.75% to 23% via SFT and DPO, yielding fine-tuned models that match or exceed closed-source VLMs on Eval-400 and ByteMorph-Bench.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"A two-stage SFT and DPO pipeline aligns vision-language models to cut critical errors in image editing instructions from 47% to 23%.","strongest_claim":"fine-tuned Qwen3-VL models outperform open-source baselines; the 235B model reaches 4.712 on Eval-400 (vs. Gemini-3-Pro 4.706, GPT-4.1 4.220, Kimi-K2.5 4.111) and 4.588 on ByteMorph-Bench (vs. Gemini-3-Pro 4.522, GPT-4.1 3.412). Human evaluation shows critical errors falling from 47.75% to 23% and correctness rising from 41.75% to 66%.","weakest_assumption":"That the three identified failure modes (orientation inconsistency, viewpoint ambiguity, insufficient fine-grained attribute description) are the dominant sources of unusable instructions and that the human preference data collected for DPO faithfully captures them without introducing new selection biases or annotation artifacts."}},"verdict_id":"1457be13-faa7-4e28-aa20-6eb82131bf13"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:7f5f23df8719843b22f1363eff7755b843ae1fd27f5e65644b89ec63d9c8bde4","target":"record","created_at":"2026-05-26T02:05:09Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"3a2e4252e971168d8233c682f68285742af2df9e05ead7a570e68d4258b92e57","cross_cats_sorted":["cs.AI"],"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CV","submitted_at":"2026-04-09T13:11:33Z","title_canon_sha256":"c82fb40e08059541fad887d45a11a27ee4e2310312c3b7193b08b4a06ec074a7"},"schema_version":"1.0","source":{"id":"2604.08213","kind":"arxiv","version":2}},"canonical_sha256":"a23dce54bad22548888c878ff5d529a7eeabbc9527d7c8edc480a0a834b81b33","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"a23dce54bad22548888c878ff5d529a7eeabbc9527d7c8edc480a0a834b81b33","first_computed_at":"2026-05-26T02:05:09.118007Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-26T02:05:09.118007Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"GjNc08To4qqRYawfAPK9hH/KWKgZ0tKkfvkg/bnzlDJfMLvLsrS8FdxMKL6+MlPkkAzmUDcHu4H21sMSIGslCQ==","signature_status":"signed_v1","signed_at":"2026-05-26T02:05:09.119027Z","signed_message":"canonical_sha256_bytes"},"source_id":"2604.08213","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:7f5f23df8719843b22f1363eff7755b843ae1fd27f5e65644b89ec63d9c8bde4","sha256:8490e0e17e3818c24cf1f6614e2baf3346f2b115e0a637a2e17db26707643cb4"],"state_sha256":"1657e4664a8c4e6b265675b1de47d6c70291983bace430333029331e368d7f5d"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"ZJZFh46U5c1/QVt6akezu/PaXYkVkB1yNAsFCjsQE98CUeY5EeeRUPdciK2Whe/8PcC2ylmizmwhaD1RO6BKBg==","signed_message":"bundle_sha256_bytes","signed_at":"2026-05-27T04:50:02.755849Z","bundle_sha256":"a7cdcb8bbc7a94e9b6c7ad8b1a41934aafa8648d7b837c8807092a0a5b767f98"}}