{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/what-makes-training-multi-modal-networks-hard","title":"What Makes Training Multi-Modal Classification Networks Hard?","arxiv_id":"1905.12681","date":"2019-05-29","proceeding":"CVPR 2020 6","authors":["Wei-Yao Wang","Du Tran","Matt Feiszli"],"abstract":"Consider end-to-end training of a multi-modal vs. a single-modal network on a task with multiple input modalities: the multi-modal network receives more information, so it should match or outperform its single-modal counterpart. In our experiments, however, we observe the opposite: the best single-modal network always outperforms the multi-modal network. This observation is consistent across different combinations of modalities and on different tasks and benchmarks. This paper identifies two main causes for this performance drop: first, multi-modal networks are often prone to overfitting due to increased capacity. Second, different modalities overfit and generalize at different rates, so training them jointly with a single optimization strategy is sub-optimal. We address these two problems with a technique we call Gradient Blending, which computes an optimal blend of modalities based on their overfitting behavior. We demonstrate that Gradient Blending outperforms widely-used baselines for avoiding overfitting and achieves state-of-the-art accuracy on various tasks including human action recognition, ego-centric action recognition, and acoustic event detection.","url_abs":"https://arxiv.org/abs/1905.12681v5","url_pdf":"https://arxiv.org/pdf/1905.12681v5.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"what-makes-training-multi-modal-networks-hard","repo_url":"https://github.com/facebookresearch/R2Plus1D","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"caffe2","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"what-makes-training-multi-modal-networks-hard","repo_url":"https://github.com/facebookresearch/VMZ","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"caffe2","reach":null},{"paper_slug":"what-makes-training-multi-modal-networks-hard","repo_url":"https://github.com/guide2157/ChulaXrayClassifier","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"none","reach":{"status":"ok"}}],"tasks":[{"task_slug":"action-classification","task_name":"Action Classification"},{"task_slug":"action-recognition-in-videos","task_name":"Action Recognition"},{"task_slug":"action-recognition-in-videos-2","task_name":"Action Recognition In Videos"},{"task_slug":"classification-1","task_name":"Classification"},{"task_slug":"event-detection","task_name":"Event Detection"},{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"multi-modal-classification","task_name":"Multi-modal Classification"},{"task_slug":"action-recognition","task_name":"Temporal Action Localization"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/action-classification-on-kinetics-400","task":"Action Classification","dataset":"Kinetics-400","model":"G-Blend (Sports-1M pretrain)","rank_in_archive_order":122,"of":207,"metrics":{"Acc@1":"78.9"},"uses_additional_data":false},{"leaderboard":"/sota/action-classification-on-kinetics-400","task":"Action Classification","dataset":"Kinetics-400","model":"G-Blend","rank_in_archive_order":135,"of":207,"metrics":{"Acc@1":"77.7"},"uses_additional_data":false},{"leaderboard":"/sota/action-recognition-in-videos-on-sports-1m-1","task":"Action Recognition In Videos","dataset":"Sports-1M","model":"G-Blend","rank_in_archive_order":1,"of":2,"metrics":{"Video hit@1":"74.8","Video hit@5":"92.4"},"uses_additional_data":false},{"leaderboard":"/sota/action-recognition-in-videos-on-minisports-1","task":"Action Recognition In Videos","dataset":"miniSports","model":"G-Blend","rank_in_archive_order":1,"of":1,"metrics":{"Clip Hit@1":"49.7","Video hit@1":"62.8","Video hit@5":"85.5"},"uses_additional_data":false}],"syntology":{"syntology_url":"https://syntology.ai/paper/1905.12681","atlas_url":"https://app.syntology.ai/?focus=1905.12681","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"1905.12681"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-25T09:33:49+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/facebookresearch/VMZ","reach":null},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/guide2157/ChulaXrayClassifier","reach":{"status":"ok"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/facebookresearch/R2Plus1D","reach":{"status":"ok","spdx":"Apache-2.0"}}],"summary":{"ran_draft_wrong":1,"unverified":10},"by_repo_kind":{"listed":{"samples":11,"ran":1,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"33fb556271040441","entry":"create_model","repo":"facebookresearch/R2Plus1D","repo_kind":"listed","path":"c2/lib/models/c3d_model.py","file_url":"https://github.com/facebookresearch/R2Plus1D/blob/HEAD/c2/lib/models/c3d_model.py","link_basis":"plan_row","language":"python","status":"ran_draft_wrong","verification_level":1,"contract_check":"OUTPUT_MISDECLARED","metamorphic_tier":"deterministic","behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"33fb556271040441"}},{"code_sha256_prefix":"dfc355d0820eea25","entry":"FlipBGR2RGB","repo":"facebookresearch/R2Plus1D","repo_kind":"listed","path":"c2/lib/utils/model_loader.py","file_url":"https://github.com/facebookresearch/R2Plus1D/blob/HEAD/c2/lib/utils/model_loader.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"dfc355d0820eea25"}},{"code_sha256_prefix":"eafa75d1a397bede","entry":"accuracy_metric","repo":"facebookresearch/R2Plus1D","repo_kind":"listed","path":"c2/lib/utils/metric.py","file_url":"https://github.com/facebookresearch/R2Plus1D/blob/HEAD/c2/lib/utils/metric.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"eafa75d1a397bede"}},{"code_sha256_prefix":"be8f28f90ea3b6fd","entry":"blobs_from_pkl","repo":"facebookresearch/R2Plus1D","repo_kind":"listed","path":"utilities/model_conversion/convert_models.py","file_url":"https://github.com/facebookresearch/R2Plus1D/blob/HEAD/utilities/model_conversion/convert_models.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"be8f28f90ea3b6fd"}},{"code_sha256_prefix":"3c5043c8ffebe8fc","entry":"ir_csn_152","repo":"facebookresearch/R2Plus1D","repo_kind":"listed","path":"utilities/model_conversion/conversion_models.py","file_url":"https://github.com/facebookresearch/R2Plus1D/blob/HEAD/utilities/model_conversion/conversion_models.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"3c5043c8ffebe8fc"}},{"code_sha256_prefix":"3f1d35c210ec7379","entry":"mean_ap_metric","repo":"facebookresearch/R2Plus1D","repo_kind":"listed","path":"c2/lib/utils/metric.py","file_url":"https://github.com/facebookresearch/R2Plus1D/blob/HEAD/c2/lib/utils/metric.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"3f1d35c210ec7379"}},{"code_sha256_prefix":"4a0058a818794147","entry":"model_validation","repo":"facebookresearch/R2Plus1D","repo_kind":"listed","path":"c2/lib/models/model_builder.py","file_url":"https://github.com/facebookresearch/R2Plus1D/blob/HEAD/c2/lib/models/model_builder.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"4a0058a818794147"}},{"code_sha256_prefix":"f5d71fa38ad9ad76","entry":"r2plus1d_152","repo":"facebookresearch/R2Plus1D","repo_kind":"listed","path":"pt/vmz/models/r2plus1d.py","file_url":"https://github.com/facebookresearch/R2Plus1D/blob/HEAD/pt/vmz/models/r2plus1d.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"f5d71fa38ad9ad76"}},{"code_sha256_prefix":"f6417808cd2990b8","entry":"r2plus1d_152","repo":"facebookresearch/R2Plus1D","repo_kind":"listed","path":"utilities/model_conversion/conversion_models.py","file_url":"https://github.com/facebookresearch/R2Plus1D/blob/HEAD/utilities/model_conversion/conversion_models.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"f6417808cd2990b8"}},{"code_sha256_prefix":"1793f9b13efb9ae4","entry":"r2plus1d_34","repo":"facebookresearch/R2Plus1D","repo_kind":"listed","path":"pt/vmz/models/r2plus1d.py","file_url":"https://github.com/facebookresearch/R2Plus1D/blob/HEAD/pt/vmz/models/r2plus1d.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"1793f9b13efb9ae4"}},{"code_sha256_prefix":"c2d08c4f7298be18","entry":"r2plus1d_34","repo":"facebookresearch/R2Plus1D","repo_kind":"listed","path":"utilities/model_conversion/conversion_models.py","file_url":"https://github.com/facebookresearch/R2Plus1D/blob/HEAD/utilities/model_conversion/conversion_models.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"c2d08c4f7298be18"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}