{"id":19146476,"url":"https://github.com/kahsolt/conv2d-kernels","last_synced_at":"2026-06-12T00:30:28.900Z","repository":{"id":113306661,"uuid":"559110057","full_name":"Kahsolt/conv2d-kernels","owner":"Kahsolt","description":"Interactive experiments on pretrained Conv2d layer weights.","archived":false,"fork":false,"pushed_at":"2023-12-18T15:46:39.000Z","size":936,"stargazers_count":1,"open_issues_count":0,"forks_count":0,"subscribers_count":2,"default_branch":"main","last_synced_at":"2025-01-03T16:28:28.791Z","etag":null,"topics":["cnn","neural-network","tutorial","visualization"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/Kahsolt.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2022-10-29T04:52:06.000Z","updated_at":"2023-12-31T04:06:23.000Z","dependencies_parsed_at":null,"dependency_job_id":"89e1c72a-4efa-45f0-8fc0-389cf1b6e741","html_url":"https://github.com/Kahsolt/conv2d-kernels","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Kahsolt%2Fconv2d-kernels","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Kahsolt%2Fconv2d-kernels/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Kahsolt%2Fconv2d-kernels/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/Kahsolt%2Fconv2d-kernels/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/Kahsolt","download_url":"https://codeload.github.com/Kahsolt/conv2d-kernels/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":240229935,"owners_count":19768588,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["cnn","neural-network","tutorial","visualization"],"created_at":"2024-11-09T07:44:14.293Z","updated_at":"2026-06-12T00:30:28.794Z","avatar_url":"https://github.com/Kahsolt.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# conv2d-kernels\n\n    Interactive experiments on pretrained Conv2d layer weights.\n\n----\n\nConv2d layers are learnable 2D signal filters, it self is indeed a function-let, so what can it actually do? 🤔\n\n### Apps\n\n#### filter\n\n⚪ inspect into featur maps: what does the **first** conv2d layer do in the well-known classifiers?\n\n![img/filter.png](img/filter.png)\n\n#### kernel\n\n⚪ inspect into conv2d kernels: what are the learned geometrical basis?\n\nThe first Conv2d layer kernels of a pretrained model:\n\n![img/kernel.png](img/kernel.png)\n\n#### cluster\n\n⚪ grouping kernels: many kernels seems to be similar thus redundant?\n\nThe first Conv2d layer kernels of a pretrained model (and centroids):\n\n| kernels | kernel centroids |\n| :-: | :-: |\n| ![img/resnet18-conv1-kernels.png](img/resnet18-conv1-kernels.png) | ![img/resnet18-conv1-kernel-centroids.png](img/resnet18-conv1-kernel-centroids.png) |\n\n#### attack\n\n⚪ PGD adversarial attack: what if we attack a single conv2d layer?\n\n![img/attack.png](img/attack.png)\n\n#### fixedpoint\n\n⚪ mathematical property of the well-known image kernels: what are the fixed points of a 2d kernel?\n\n![img/fixedpoint.png](img/fixedpoint.png)\n\n\n#### resources download\n\n- The reprocessed ImageNet-1k dataset can be downloaded here: [https://pan.quark.cn/s/373b488d101e](https://pan.quark.cn/s/373b488d101e)\n  - NOTE: It is a subset of 1k images from validation split of original intact ImageNet dataset\n- Tiny-ImageNet can be found here: [tiny-imagenet-200](https://tiny-imagenet.herokuapp.com)\n\n----\n\nby Armit\n2022/10/28 \n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fkahsolt%2Fconv2d-kernels","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fkahsolt%2Fconv2d-kernels","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fkahsolt%2Fconv2d-kernels/lists"}