feat(YOLOX): update README and requirements.txt

README.md

@@ -3,11 +3,18 @@
 
 ## Introduction
 YOLOX is an anchor-free version of YOLO, with a simpler design but better performance! It aims to bridge the gap between research and industrial communities.
+For more details, please refer to our [Arxiv report]().
 
 <img src="assets/git_fig.png" width="1000" >
 
 ## Updates!!
-* 【2020/07/19】 We have released our technical report on Arxiv.
+* 【2020/07/19】 We have released our technical report on [Arxiv]().
+
+## Comming soon
+- [ ] YOLOX-P6 and larger model.
+- [ ] Obj365 pretrain.
+- [ ] Transformer modules.
+- [ ] More features in need.
 
 ## Benchmark
 
@@ -31,22 +38,29 @@ YOLOX is an anchor-free version of YOLO, with a simpler design but better perfor
 <details>
 <summary>Installation</summary>
 
-Step1. Install [apex](https://github.com/NVIDIA/apex).
+Step1. Install YOLOX.
+```shell
+git clone [email protected]:Megvii-BaseDetection/YOLOX.git
+cd yolox
+pip3 install -U pip && pip3 install -v -e .  # or  python3 setup.py develop
+```
+Step2. Install [apex](https://github.com/NVIDIA/apex).
 
 ```shell
 git clone https://github.com/NVIDIA/apex
 cd apex
 pip3 install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./
 ```
-Step2. Install YOLOX.
-```bash
-$ git clone [email protected]:Megvii-BaseDetection/YOLOX.git
-$ cd yolox
-$ pip3 install -v -e .  # or "python3 setup.py develop
+Step3. Install [pycocotools](https://github.com/cocodataset/cocoapi).
+
+```shell
+pip3 install cython; pip3 install 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'
 ```
 
 </details>
 
+</details>
+
 <details>
 <summary>Demo</summary>
 

demo/TensorRT/cpp/README.md

@@ -8,6 +8,8 @@ our C++ demo will not include the model converting or constructing like other te
 
 Follow the trt [python demo README](../python/README.md) to convert and save the serialized engine file.
 
+Check the 'model_trt.engine' file generated from Step 1, which will automatically saved at the current demo dir.
+
 
 ## Step 2: build the demo
 
@@ -24,20 +26,14 @@ cmake ..
 make
 ```
 
-Move the 'model_trt.engine' file generated from Step 1 (saved at the exp output dir) to the build dir:
-
-```shell
-mv /path/to/your/exp/output/dir/model_trt.engine .
-```
-
 Then run the demo:
 
 ```shell
-./yolox -d /your/path/to/yolox/assets
+./yolox ../model_trt.engine -i ../../../../assets/dog.jpg
 ```
 
 or
 
 ```shell
-./yolox -d <img dir>
+./yolox <path/to/your/engine_file> -i <path/to/image>
 ```

demo/TensorRT/cpp/yolox.cpp

@@ -230,25 +230,6 @@ float* blobFromImage(cv::Mat& img){
 }
 
 
-int read_files_in_dir(const char *p_dir_name, std::vector<std::string> &file_names) {
-    DIR *p_dir = opendir(p_dir_name);
-    if (p_dir == nullptr) {
-        return -1;
-    }
-
-    struct dirent* p_file = nullptr;
-    while ((p_file = readdir(p_dir)) != nullptr) {
-        if (strcmp(p_file->d_name, ".") != 0 &&
-                strcmp(p_file->d_name, "..") != 0) {
-            std::string cur_file_name(p_file->d_name);
-            file_names.push_back(cur_file_name);
-        }
-    }
-
-    closedir(p_dir);
-    return 0;
-}
-
 static void decode_outputs(float* prob, std::vector<Object>& objects, float scale, const int img_w, const int img_h) {
         std::vector<Object> proposals;
         std::vector<int> strides = {8, 16, 32};
@@ -432,7 +413,7 @@ static void draw_objects(const cv::Mat& bgr, const std::vector<Object>& objects,
                     cv::FONT_HERSHEY_COMPLEX, 0.4, txt_color, 1);
     }
 
-    cv::imwrite("_" + f, image);
+    cv::imwrite("det_res.jpg", image);
     fprintf(stderr, "save vis file\n");
     /* cv::imshow("image", image); */
     /* cv::waitKey(0); */
@@ -482,8 +463,9 @@ int main(int argc, char** argv) {
     char *trtModelStream{nullptr};
     size_t size{0};
 
-    if (argc == 3 && std::string(argv[1]) == "-d") {
-        std::ifstream file("model_trt.engine", std::ios::binary);
+    if (argc == 4 && std::string(argv[2]) == "-i") {
+        const std::string engine_file_path {argv[1]};
+        std::ifstream file(engine_file_path, std::ios::binary);
         if (file.good()) {
             file.seekg(0, file.end);
             size = file.tellg();
@@ -496,15 +478,17 @@ int main(int argc, char** argv) {
     } else {
         std::cerr << "arguments not right!" << std::endl;
         std::cerr << "run 'python3 yolox/deploy/trt.py -n yolox-{tiny, s, m, l, x}' to serialize model first!" << std::endl;
-        std::cerr << "./yolox -d ../samples  // deserialize file and run inference" << std::endl;
+        std::cerr << "Then use the following command:" << std::endl;
+        std::cerr << "./yolox ../model_trt.engine -i ../../../assets/dog.jpg  // deserialize file and run inference" << std::endl;
         return -1;
     }
+    const std::string input_image_path {argv[3]};
 
-    std::vector<std::string> file_names;
-    if (read_files_in_dir(argv[2], file_names) < 0) {
-        std::cout << "read_files_in_dir failed." << std::endl;
-        return -1;
-    }
+    //std::vector<std::string> file_names;
+    //if (read_files_in_dir(argv[2], file_names) < 0) {
+        //std::cout << "read_files_in_dir failed." << std::endl;
+        //return -1;
+    //}
 
     IRuntime* runtime = createInferRuntime(gLogger);
     assert(runtime != nullptr);
@@ -520,33 +504,27 @@ int main(int argc, char** argv) {
     }
     static float* prob = new float[output_size];
 
-    int fcount = 0;
-    for (auto f: file_names) {
-        fcount++;
-        std::cout << fcount << "  " << f << std::endl;
-        cv::Mat img = cv::imread(std::string(argv[2]) + "/" + f);
-        if (img.empty()) continue;
-        int img_w = img.cols;
-        int img_h = img.rows;
-        cv::Mat pr_img = static_resize(img);
-        std::cout << "blob image" << std::endl;
-
-        float* blob;
-        blob = blobFromImage(pr_img);
-        float scale = std::min(INPUT_W / (img.cols*1.0), INPUT_H / (img.rows*1.0));
-
-        // Run inference
-        auto start = std::chrono::system_clock::now();
-        doInference(*context, blob, prob, output_size, pr_img.size());
-        auto end = std::chrono::system_clock::now();
-        std::cout << std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count() << "ms" << std::endl;
-
-        std::vector<Object> objects;
-        decode_outputs(prob, objects, scale, img_w, img_h);
-        draw_objects(img, objects, f);
-    }
+    cv::Mat img = cv::imread(input_image_path);
+    int img_w = img.cols;
+    int img_h = img.rows;
+    cv::Mat pr_img = static_resize(img);
+    std::cout << "blob image" << std::endl;
+
+    float* blob;
+    blob = blobFromImage(pr_img);
+    float scale = std::min(INPUT_W / (img.cols*1.0), INPUT_H / (img.rows*1.0));
+
+    // run inference
+    auto start = std::chrono::system_clock::now();
+    doInference(*context, blob, prob, output_size, pr_img.size());
+    auto end = std::chrono::system_clock::now();
+    std::cout << std::chrono::duration_cast<std::chrono::milliseconds>(end - start).count() << "ms" << std::endl;
+
+    std::vector<Object> objects;
+    decode_outputs(prob, objects, scale, img_w, img_h);
+    draw_objects(img, objects, input_image_path);
 
-    // Destroy the engine
+    // destroy the engine
     context->destroy();
     engine->destroy();
     runtime->destroy();

demo/TensorRT/python/README.md

@@ -37,10 +37,10 @@ The converted model and the serialized engine file (for C++ demo) will be saved
 The TensorRT python demo is merged on our pytorch demo file, so you can run the pytorch demo command with ```--trt```.
 
 ```shell
-python tools/demo.py -n yolox-s --trt --conf 0.3 --nms 0.65 --tsize 640
+python tools/demo.py image -n yolox-s --trt --save_result
 ```
 or
 ```shell
-python tools/demo.py -f exps/base/yolox_s.py --trt --conf 0.3 --nms 0.65 --tsize 640
+python tools/demo.py image -f exps/base/yolox_s.py --trt --save_result
 ```
 

requirements.txt

@@ -4,14 +4,11 @@ opencv_python
 loguru
 scikit_image
 tqdm
-apex
 torchvision
-pycocotools
-apex
 Pillow
 skimage
 thop
 ninja
 tabulate
 tensorboard
-onnxruntime
+onnxruntime

setup.py

@@ -47,6 +47,10 @@ with open("yolox/__init__.py", "r") as f:
     ).group(1)
 
 
+with open("requirements.txt", "r") as f:
+    reqs = [x.strip() for x in f.readlines()]
+
+
 with open("README.md", "r") as f:
     long_description = f.read()
 
@@ -59,6 +63,7 @@ setuptools.setup(
     long_description=long_description,
     ext_modules=get_extensions(),
     classifiers=["Programming Language :: Python :: 3", "Operating System :: OS Independent"],
+    install_requires=reqs,
     cmdclass={"build_ext": torch.utils.cpp_extension.BuildExtension},
     packages=setuptools.find_packages(),
 )

tools/demo.py

@@ -15,7 +15,7 @@ import torch.backends.cudnn as cudnn
 from yolox.data.data_augment import preproc
 from yolox.data.datasets import COCO_CLASSES
 from yolox.exp import get_exp
-from yolox.utils import fuse_model, get_model_info, postprocess, setup_logger, vis, xyxy2xywh
+from yolox.utils import fuse_model, get_model_info, postprocess, setup_logger, vis
 
 IMAGE_EXT = ['.jpg', '.jpeg', '.webp', '.bmp', '.png']
 
@@ -26,7 +26,7 @@ def make_parser():
     parser.add_argument("-expn", "--experiment-name", type=str, default=None)
     parser.add_argument("-n", "--name", type=str, default=None, help="model name")
 
-    parser.add_argument('--path', default='./demo', help='path to images or video')
+    parser.add_argument('--path', default='./assets/dog.jpg', help='path to images or video')
     parser.add_argument('--camid', type=int, default=0, help='webcam demo camera id')
     parser.add_argument(
         '--save_result', action='store_true',

tools/trt.py

@@ -64,7 +64,7 @@ def main():
     torch.save(model_trt.state_dict(), os.path.join(file_name, 'model_trt.pth'))
     logger.info("Converted TensorRT model done.")
     engine_file = os.path.join(file_name, 'model_trt.engine')
-    engine_file_demo = os.path.join('yolox', 'deploy', 'demo_trt_c++', 'model_trt.engine')
+    engine_file_demo = os.path.join('demo', 'TensorRT', 'cpp', 'model_trt.engine')
     with open(engine_file, 'wb') as f:
         f.write(model_trt.engine.serialize())