--- license: other license_name: syncfix-model-and-runtime-terms license_link: https://huggingface.co/agents2agents/SyncFix-RTX-5090/blob/main/LICENSE pipeline_tag: image-to-image tags: - tensorrt - rtx-5090 - syncfix --- # SyncFix for RTX 5090 Process three degraded views and one reference image to produce three corrected views. Inputs must be RGB and may have different dimensions. Each output has the same width and height as its corresponding degraded input image. ## Requirements - Linux x86_64 with glibc 2.35 or newer, and an NVIDIA GeForce RTX 5090 with 32 GB memory. - NVIDIA driver 580.178.04 or newer; one visible GPU. - Python 3.12, PyTorch 2.10.0+cu128, Triton 3.6.0, NumPy 1.26.4, Pillow 11.3.0 and safetensors 0.6.2. - TensorRT 11.3.0 build 99 for CUDA 12.9, including `libnvinfer.so.11` and `libnvinfer_plugin.so.11`. Use the specified software versions. Compatible newer NVIDIA drivers are supported; a matching system CUDA toolkit is not required. If TensorRT is not on the library search path, pass `--trt-lib-dir /path/to/TensorRT/lib`. Obtain the specified runtime from [NVIDIA TensorRT downloads](https://developer.nvidia.com/tensorrt). On a machine with multiple GPUs, select one with `CUDA_VISIBLE_DEVICES`. ## Install and test Download the package with the [Hugging Face CLI](https://huggingface.co/docs/huggingface_hub/guides/cli): ```bash hf download agents2agents/SyncFix-RTX-5090 --local-dir syncfix-rtx5090 cd syncfix-rtx5090 ``` ```bash bash setup.sh source .venv/bin/activate python verify.py python verify.py --run --trt-lib-dir /path/to/TensorRT/lib ``` `verify.py` checks the downloaded files using Python alone. `--run` also checks inference using the included example and a group of images with different dimensions. Numerical checks allow small variation between processes. The first inference includes initialization and warmup. ## Run ```bash python run.py \ --degraded example/degraded_0.png example/degraded_1.png example/degraded_2.png \ --reference example/reference.png --seed 42 \ --output-dir /path/to/new-results \ --trt-lib-dir /path/to/TensorRT/lib ``` Choose a new or empty output directory. The command saves three PNGs and a request record. Supplying a seed resets sampling for that request; omitting it continues the current random state. For multiple groups, save a JSONL file: ```json {"id":"group000","degraded":["a.png","b.png","c.png"],"reference":"ref.png","seed":42} {"id":"group001","degraded":["d.png","e.png","f.png"],"reference":"ref.png"} ``` ```bash python run.py --requests /data/scene.jsonl --output-dir /data/corrected ``` Image paths resolve relative to the request file. A JSON array is also accepted. The model remains loaded across requests. ## Python ```python from PIL import Image from syncfix_5090 import Pipeline paths = ["a.png", "b.png", "c.png", "reference.png"] images = [Image.open(path).convert("RGB") for path in paths] with Pipeline("/path/to/bundle", trt_lib_dir="/path/to/TensorRT/lib") as model: corrected = model.infer(images[:3], images[3], seed=42) ``` Create one pipeline per process, before initializing CUDA elsewhere. Use it from its creating thread. Returned images remain valid after later requests. See [LICENSE](https://huggingface.co/agents2agents/SyncFix-RTX-5090/blob/main/LICENSE) and [NOTICE](https://huggingface.co/agents2agents/SyncFix-RTX-5090/blob/main/NOTICE) for applicable terms and attribution.