| #include "clip.h" |
| #include "cmdline.hpp" |
| #include "timer.hpp" |
| #include <fstream> |
| #include <cstring> |
| #include <opencv2/opencv.hpp> |
|
|
| int main(int argc, char *argv[]) |
| { |
| clip_devices_t clip_devices; |
| memset(&clip_devices, 0, sizeof(clip_devices_t)); |
| if (clip_enum_devices(&clip_devices) != 0) |
| { |
| printf("enum devices failed\n"); |
| return -1; |
| } |
|
|
| if (clip_devices.host.available) |
| { |
| clip_sys_init(host_device, -1); |
| } |
| else if (clip_devices.devices.count > 0) |
| { |
| clip_sys_init(axcl_device, 0); |
| } |
| else |
| { |
| printf("no device available\n"); |
| return -1; |
| } |
|
|
| clip_init_t init_info; |
| memset(&init_info, 0, sizeof(init_info)); |
|
|
| cmdline::parser parser; |
| parser.add<std::string>("ienc", 0, "encoder model(onnx model or axmodel)", true, "cnclip/cnclip_vit_l14_336px_vision_u16u8.axmodel"); |
| parser.add<std::string>("tenc", 0, "text encoder model(onnx model or axmodel)", true, "cnclip/cnclip_vit_l14_336px_text_u16.axmodel"); |
| parser.add<std::string>("vocab", 'v', "vocab path", true, "cnclip/cn_vocab.txt"); |
| parser.add<int>("language", 'l', "language choose, 0:english 1:chinese", false, 1); |
| parser.add<std::string>("db_path", 'd', "db path", false, "clip_feat_db"); |
|
|
| parser.add<std::string>("image", 'i', "image folder(jpg png etc....)", true); |
| parser.add<std::string>("text", 't', "text or txt file", true); |
| parser.parse_check(argc, argv); |
|
|
| sprintf(init_info.image_encoder_path, "%s", parser.get<std::string>("ienc").c_str()); |
| sprintf(init_info.text_encoder_path, "%s", parser.get<std::string>("tenc").c_str()); |
| sprintf(init_info.tokenizer_path, "%s", parser.get<std::string>("vocab").c_str()); |
| init_info.isCN = parser.get<int>("language"); |
| sprintf(init_info.db_path, "%s", parser.get<std::string>("db_path").c_str()); |
|
|
| printf("image_encoder_path: %s\n", init_info.image_encoder_path); |
| printf("text_encoder_path: %s\n", init_info.text_encoder_path); |
| printf("tokenizer_path: %s\n", init_info.tokenizer_path); |
| printf("isCN: %d\n", init_info.isCN); |
| printf("db_path: %s\n", init_info.db_path); |
|
|
| if (clip_devices.host.available) |
| { |
| init_info.dev_type = host_device; |
| } |
| else if (clip_devices.devices.count > 0) |
| { |
| init_info.dev_type = axcl_device; |
| init_info.devid = 0; |
| } |
|
|
| clip_handle_t handle; |
| int ret = clip_create(&init_info, &handle); |
| if (ret != clip_errcode_success) |
| { |
| printf("clip_create failed\n"); |
| return -1; |
| } |
|
|
| std::string image_src = parser.get<std::string>("image"); |
| std::string text = parser.get<std::string>("text"); |
|
|
| std::vector<std::string> image_paths; |
| cv::glob(image_src + "/*.*", image_paths); |
|
|
| for (size_t i = 0; i < image_paths.size(); i++) |
| { |
| std::string image_path = image_paths[i]; |
| std::string image_name = image_path.substr(image_path.find_last_of("/") + 1); |
| char key[CLIP_KEY_MAX_LEN]; |
| sprintf(key, "%s", image_name.c_str()); |
| if (clip_contain(handle, key)) |
| { |
| |
| continue; |
| } |
|
|
| cv::Mat src = cv::imread(image_path); |
| cv::cvtColor(src, src, cv::COLOR_BGR2RGB); |
| clip_image_t image; |
| image.data = src.data; |
| image.width = src.cols; |
| image.height = src.rows; |
| image.channels = src.channels(); |
| image.stride = src.step; |
|
|
| timer t; |
| clip_add(handle, key, &image, 0); |
| |
| } |
| int topk = 10; |
| std::vector<clip_result_item_t> results(topk); |
| timer t; |
| clip_match_text(handle, text.c_str(), results.data(), topk); |
| printf("match text \"%s\" %6.2fms\n", text.c_str(), t.cost()); |
| printf("|%32s | %6s|\n", "key", "score"); |
| for (size_t i = 0; i < results.size(); i++) |
| { |
| printf("|%32s | %6.2f|\n", results[i].key, results[i].score); |
| } |
|
|
| clip_destroy(handle); |
|
|
| if (clip_devices.host.available) |
| { |
| clip_sys_deinit(host_device, -1); |
| } |
| else if (clip_devices.devices.count > 0) |
| { |
|
|
| clip_sys_deinit(axcl_device, 0); |
| } |
|
|
| return 0; |
| } |