README.md

@@ -30,37 +30,15 @@ while a model downloads and loads.
 
 ### Web UI
 
-Open **http://localhost:8000/** — a single-page test app (handy over SSH):
+Open **http://localhost:8000/** — a two-tab test page (handy over SSH):
 
 - **Auto remove** — pick a model variant + resolution.
 - **Prompt segment** — type what to keep (e.g. `the dog`), tune the
   GroundingDINO box / text thresholds.
 
-Both modes support a transparency checkerboard preview, click-to-zoom lightbox,
+Both tabs support a transparency checkerboard preview, click-to-zoom lightbox,
 optional crop-to-subject, and download.
 
-#### Keyboard shortcuts
-
-The UI is fully keyboard-drivable. Shortcuts are ignored while typing in a
-field and while Ctrl/Cmd/Alt is held.
-
-| Key                 | Action                                        |
-|---------------------|-----------------------------------------------|
-| `B`                 | Toggle the controls sidebar                   |
-| `U`                 | Open the file picker to upload an image       |
-| `I` / `O`           | Show the input / output image                 |
-| `F` / `Z`           | Open the zoom view for the visible image      |
-| `S`                 | Save (download PNG), once a result exists     |
-
-In the zoom view:
-
-| Key                       | Action                                  |
-|---------------------------|-----------------------------------------|
-| `F` / `Z` / `Esc`         | Close the zoom view                     |
-| `+` / `-`                 | Zoom in / out (1×–8×)                   |
-| `0`                       | Reset zoom & pan                        |
-| Arrows or `H` `J` `K` `L` | Pan (while zoomed past 1×)              |
-
 ## API
 
 ### `POST /predict` — automatic background removal
@@ -130,7 +108,7 @@ make dev        # uv sync + run the server locally
 src/rmbg_as_a_service/model.py           BiRefNet / RMBG-2.0 wrapper + compositing
 src/rmbg_as_a_service/prompt_segment.py  GroundingDINO + SAM pipeline
 src/rmbg_as_a_service/server.py          LitServe /predict + /segment + web UI
-src/rmbg_as_a_service/static/            web UI (index.html + styles.css)
+src/rmbg_as_a_service/static/            web UI (index.html)
 scripts/client.py                       stdlib-only test client
 Dockerfile / compose.yml                CUDA image + nvidia runtime
 Makefile                                build / run / test shortcuts

build/Dockerfile

@@ -1,32 +1,12 @@
-# BiRefNet background removal service — installs the published package on the CUDA 12.6 runtime.
+FROM python:3.13-slim
-# Same CUDA base family as the main ../Dockerfile, but installs rmbg-as-a-service from TestPyPI
-# with a plain pip rather than building from the local source tree.
-FROM nvidia/cuda:12.6.1-cudnn-runtime-ubuntu24.04
 
 ENV DEBIAN_FRONTEND=noninteractive \
     PYTHONUNBUFFERED=1 \
-    PIP_BREAK_SYSTEM_PACKAGES=1 \
     HF_HOME=/app/hf_cache \
     HF_HUB_ENABLE_HF_TRANSFER=1 \
     PORT=8000
 
-# Ubuntu 24.04 ships Python 3.12 (the project requires >=3.12).
+RUN pip install rmbg-as-a-service hf-transfer
-RUN apt-get update \
-    && apt-get install -y --no-install-recommends python3 python3-pip ca-certificates \
-    && rm -rf /var/lib/apt/lists/*
-
-# Fetch only the rmbg-as-a-service wheel from TestPyPI (--no-deps), so TestPyPI never takes
-# part in dependency resolution -- otherwise junk squatted packages there (e.g. "FASTAPI")
-# outrank the real ones. Then install it with real PyPI as the index, CUDA torch from the
-# PyTorch index (its +cu126 build outranks the plain wheel by local-version ordering).
-RUN --mount=type=cache,target=/root/.cache/pip \
-    pip download --no-deps --dest /tmp/pkg \
-        --index-url https://test.pypi.org/simple/ \
-        "rmbg-as-a-service==0.0.2" \
-    && pip install \
-        --extra-index-url https://download.pytorch.org/whl/cu126 \
-        /tmp/pkg/*.whl hf-transfer \
-    && rm -rf /tmp/pkg
 
 EXPOSE 8000
 CMD ["rmbg-as-a-service"]

build/Makefile

@@ -1,53 +0,0 @@
-# mindthemath/rmbg image — build the published-package container, test it, push it.
-# Override inline, e.g.:  make release TAG=0.0.2
-
-COMPOSE ?= docker compose
-PYTHON  ?= python3
-IMAGE   ?= mindthemath/rmbg
-TAG     ?= latest
-PORT    ?= 8000
-INPUT   ?= ../test.jpg
-OUTPUT  ?= output.png
-BG      ?= alpha
-
-# Exported so compose.yml's ${TAG} interpolation picks it up.
-export TAG
-
-.DEFAULT_GOAL := help
-
-.PHONY: help build run up stop down logs log ps shell test push release clean
-
-help: ## Show this help
-	@grep -E '^[a-zA-Z_-]+:.*?## ' $(MAKEFILE_LIST) \
-		| awk 'BEGIN{FS=":.*?## "}{printf "  \033[36m%-12s\033[0m %s\n", $$1, $$2}'
-
-build: ## Build the mindthemath/rmbg image
-	$(COMPOSE) build
-
-run up: ## Start the service (GPU) in the background
-	$(COMPOSE) up -d
-
-stop down: ## Stop and remove the service container
-	$(COMPOSE) down
-
-logs log: ## Follow service logs
-	$(COMPOSE) logs -f
-
-ps: ## Show service status
-	$(COMPOSE) ps
-
-shell: ## Open a shell inside a fresh container
-	$(COMPOSE) run --rm --entrypoint bash rmbg
-
-test: ## Send INPUT to the running service, save OUTPUT
-	$(PYTHON) ../scripts/client.py --url http://localhost:$(PORT) \
-		--input $(INPUT) --output $(OUTPUT) --background $(BG)
-
-push: ## Push mindthemath/rmbg:$(TAG) to Docker Hub (needs docker login)
-	docker push $(IMAGE):$(TAG)
-
-release: build push ## Build then push mindthemath/rmbg:$(TAG)
-
-clean: ## Stop the service and remove the built image
-	-$(COMPOSE) down
-	-docker image rm $(IMAGE):$(TAG)

build/compose.yml

@@ -1,31 +0,0 @@
-# mindthemath/rmbg — runs the image built from the published (TestPyPI) package.
-services:
-  rmbg:
-    build:
-      context: .
-      dockerfile: Dockerfile
-    image: mindthemath/rmbg:${TAG:-latest}
-    container_name: rmbg
-    ports:
-      - "${PORT:-8001}:8000"
-    environment:
-      - NVIDIA_VISIBLE_DEVICES=all
-      - NVIDIA_DRIVER_CAPABILITIES=compute,utility
-      # Default variant/resolution; both are also selectable per request.
-      - BIREFNET_MODEL=${BIREFNET_MODEL:-general}
-      - BIREFNET_RESOLUTION=${BIREFNET_RESOLUTION:-1024}
-    # Use the nvidia-container-runtime for GPU acceleration.
-    runtime: nvidia
-    volumes:
-      # Persist downloaded BiRefNet weights across container restarts.
-      - hf-cache:/app/hf_cache
-    healthcheck:
-      test: ["CMD", "python3", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:8000/health')"]
-      interval: 15s
-      timeout: 5s
-      retries: 30
-      start_period: 180s
-    restart: unless-stopped
-
-volumes:
-  hf-cache:

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "rmbg-as-a-service"
-version = "0.0.3dev1"
+version = "0.0.2"
 description = "Background removal as a GPU-accelerated API"
 readme = "README.md"
 requires-python = ">=3.12"

src/rmbg_as_a_service/static/index.html

@@ -161,7 +161,7 @@
 <!-- lightbox -->
 <div id="lightbox" class="lightbox" hidden>
   <div class="lb-bar">
-    <span>SCROLL / +&minus; ZOOM &middot; DRAG OR HJKL / ARROWS PAN &middot; 0 RESETS &middot; F / Z / ESC CLOSE</span>
+    <span>SCROLL TO ZOOM &middot; DRAG TO PAN &middot; DOUBLE-CLICK RESETS &middot; ESC CLOSES</span>
     <button class="lb-close" id="lbClose" title="Close">&#10005;</button>
   </div>
   <div class="lb-stage" id="lbStage"><img id="lbImg" alt="" /></div>
@@ -350,39 +350,6 @@ function openLightbox(src, isResult) {
 }
 function closeLightbox() { lightbox.hidden = true; lbImg.removeAttribute('src'); }
 
-/* which preview image is currently shown */
-function currentImg() {
-  return tabOutput.classList.contains('active') ? outImg : srcImg;
-}
-/* 'f' / 'z' — open the zoom view for the visible image, or close it */
-function toggleZoom() {
-  if (!lightbox.hidden) { closeLightbox(); return; }
-  const img = currentImg();
-  const src = img.getAttribute('src');
-  if (src) openLightbox(src, img === outImg);
-}
-/* keyboard zoom — anchored on the centre of the stage */
-function lbZoom(factor) {
-  const stageRect = lbStage.getBoundingClientRect();
-  const imgRect = lbImg.getBoundingClientRect();
-  const cx = stageRect.left + stageRect.width / 2 - imgRect.left;
-  const cy = stageRect.top + stageRect.height / 2 - imgRect.top;
-  const newScale = Math.min(8, Math.max(1, lbScale * factor));
-  const ratio = newScale / lbScale;
-  lbTx -= cx * (ratio - 1);
-  lbTy -= cy * (ratio - 1);
-  lbScale = newScale;
-  if (lbScale === 1) { lbTx = 0; lbTy = 0; }
-  lbApply();
-}
-/* keyboard pan — dx/dy in {-1,0,1}; only meaningful while zoomed in */
-function lbPan(dx, dy) {
-  if (lbScale <= 1) return;
-  lbTx += dx * 80;
-  lbTy += dy * 80;
-  lbApply();
-}
-
 srcImg.addEventListener('click', () => openLightbox(srcImg.getAttribute('src'), false));
 outImg.addEventListener('click', () => openLightbox(outImg.getAttribute('src'), true));
 lbClose.addEventListener('click', closeLightbox);
@@ -390,33 +357,14 @@ lightbox.addEventListener('mousedown', e => {
   if (e.target === lightbox || e.target === lbStage) closeLightbox();
 });
 document.addEventListener('keydown', e => {
+  if (e.key === 'Escape' && !lightbox.hidden) closeLightbox();
   if (e.metaKey || e.ctrlKey || e.altKey) return;
-  const k = e.key.toLowerCase();
-
-  // --- zoom view: capture all navigation keys while it's open ---
-  if (!lightbox.hidden) {
-    if (e.key === 'Escape' || k === 'f' || k === 'z') closeLightbox();
-    else if (k === '+' || k === '=') lbZoom(1.25);
-    else if (k === '-' || k === '_') lbZoom(1 / 1.25);
-    else if (k === '0') lbReset();
-    else if (e.key === 'ArrowLeft'  || k === 'h') lbPan( 1,  0);
-    else if (e.key === 'ArrowRight' || k === 'l') lbPan(-1,  0);
-    else if (e.key === 'ArrowUp'    || k === 'k') lbPan( 0,  1);
-    else if (e.key === 'ArrowDown'  || k === 'j') lbPan( 0, -1);
-    else return;
-    e.preventDefault();
-    return;
-  }
-
-  // --- main UI shortcuts (ignored while typing in a field) ---
   const t = e.target;
   if (t && (t.tagName === 'INPUT' || t.tagName === 'TEXTAREA' || t.tagName === 'SELECT')) return;
+  const k = e.key.toLowerCase();
   if (k === 'b') toggleSidebar();
-  else if (k === 'u') fileInput.click();
   else if (k === 'i') showView('input');
   else if (k === 'o') showView('output');
-  else if (k === 'f' || k === 'z') toggleZoom();
-  else if (k === 's' && !dlbtn.disabled) dl.click();
 });
 
 lbStage.addEventListener('wheel', e => {

uv.lock

@@ -1163,7 +1163,7 @@ wheels = [
 
 [[package]]
 name = "rmbg-as-a-service"
-version = "0.0.3.dev1"
+version = "0.0.2"
 source = { editable = "." }
 dependencies = [
     { name = "einops" },