{"url":"/method/condconv","slug":"condconv","name":"CondConv","full_name":"CondConv","full_name_withheld":false,"description_markdown":"**CondConv**, or **Conditionally Parameterized Convolutions**, are a type of [convolution](https://paperswithcode.com/method/convolution) which learn specialized convolutional kernels for each example. In particular, we parameterize the convolutional kernels in a CondConv layer as a linear combination of $n$ experts $(\\alpha_1 W_1 + \\ldots + \\alpha_n W_n) * x$, where $\\alpha_1, \\ldots, \\alpha_n$ are functions of the input learned through gradient descent. To efficiently increase the capacity of a CondConv layer, developers can increase the number of experts. This can be more computationally efficient than increasing the size of the convolutional kernel itself, because the convolutional kernel is applied at many different positions within the input, while the experts are combined only once per input.","description_state":"present","introduced_year":null,"introduced_by":{"title":"CondConv: Conditionally Parameterized Convolutions for Efficient Inference","paper":"/paper/soft-conditional-computation","first_author":"Brandon Yang","n_authors":4,"url_abs":null,"archive_paper_url":"https://paperswithcode.com/paper/soft-conditional-computation"},"source":{"url":"https://arxiv.org/abs/1904.04971v3","title":"CondConv: Conditionally Parameterized Convolutions for Efficient Inference","url_on_a_paper_host":true},"code_snippet_url":"https://github.com/tensorflow/tpu/blob/bb03c18d7c2501b4df070c9936a46a9dcd6ad1cb/models/official/efficientnet/condconv/condconv_layers.py#L63","code_snippet_url_on_a_code_host":true,"categories":[{"area":"Computer Vision","area_id":"computer-vision","collection":"Convolutions","url":"/methods/category/convolutions","pwc_aliases":[]}],"n_papers_tagged":5,"archive_num_papers":5,"papers_newest_first":[{"paper":"/paper/frequency-dynamic-convolution-for-dense-image","title":"Frequency Dynamic Convolution for Dense Image Prediction","date":"2025-03-24","arxiv_id":"2503.18783","n_code_links":1,"syntology":{"ran":0,"of":2,"unverified":2,"pointer_only":0}},{"paper":"/paper/collaboration-of-experts-achieving-80-top-1","title":"Collaboration of Experts: Achieving 80% Top-1 Accuracy on ImageNet with 100M FLOPs","date":"2021-07-08","arxiv_id":"2107.03815","n_code_links":0,"syntology":null},{"paper":"/paper/extending-conditional-convolution-structures","title":"EXTENDING CONDITIONAL CONVOLUTION STRUCTURES FOR ENHANCING MULTITASKING CONTINUAL LEARNING","date":"2020-12-07","arxiv_id":null,"n_code_links":1,"syntology":null},{"paper":"/paper/weightnet-revisiting-the-design-space-of","title":"WeightNet: Revisiting the Design Space of Weight Networks","date":"2020-07-23","arxiv_id":"2007.11823","n_code_links":2,"syntology":{"ran":1,"of":1,"unverified":0,"pointer_only":0}},{"paper":"/paper/soft-conditional-computation","title":"CondConv: Conditionally Parameterized Convolutions for Efficient Inference","date":"2019-04-10","arxiv_id":"1904.04971","n_code_links":9,"syntology":null}],"papers_shown":5,"tasks":[{"task":"/task/image-classification","name":"Image Classification","papers":2},{"task":"/task/object-detection","name":"Object Detection","papers":2},{"task":"/task/computational-efficiency","name":"Computational Efficiency","papers":1},{"task":"/task/continual-learning","name":"Continual Learning","papers":1},{"task":"/task/classification","name":"General Classification","papers":1},{"task":"/task/multi-task-learning","name":"Multi-Task Learning","papers":1},{"task":"/task/prediction","name":"Prediction","papers":1},{"task":"/task/object-detection-1","name":"object-detection","papers":1}],"tasks_shown":8,"n_tasks":8,"usage_by_year":[{"year":"2019","papers":1},{"year":"2020","papers":2},{"year":"2021","papers":1},{"year":"2025","papers":1}],"row_source":"methods_table","archive":{"source":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","archive_url":"https://paperswithcode.com/method/condconv"},"syntology_read_at":"2026-09-24T18:15:14+00:00"}